“Can you all see my screen?” asks the founder, nervously. A couple of people give a thumbs up, and off she goes with her pitch. The problem is urgent, and yet no one has approached it this way — yet. The solution is scalable, the market is substantial (and growing), and the path to hockey-stick growth is credible.

Listening to startups pitch their vision is great fun. As angel investors, we love the energy, the passion, the creativity. But we also know that the weakest slide in the deck will be the one about competition.

“So what’s stopping a larger company from copying you?” is usually the first question to come up.

Very few founders can explain convincingly why a well-resourced competitor couldn’t replicate their idea. They design a castle in exhaustive detail, but pay little attention to the moat that determines whether that castle survives or not.

Scale-ups and established SMEs face a similar challenge. The walls may be higher — brand, customer relationships, operational depth — but the same barbarians are battering at the gates, and they have very deep pockets.

Barbarians at the gate

There is a pattern in how Anthropic, OpenAI and Google expand their footprint. A capability first emerges in research, then becomes a product feature and finally a direct market entry that can render whole categories of software redundant.

Claude is the most visible example. Anthropic’s coding agent didn’t just add code generation as a feature. It entered the software development process directly, threatening dozens of startups that built their castles around these workflows.

The same dynamic is now playing out in design, where AI-native tools are absorbing layout, asset creation, and UX work that once required both specialist skills and specialist software. In financial services and healthcare, the Labs are moving into analysis, reporting, and compliance, charging into territories hitherto occupied by specialist firms.

Which sector is next? Legal is already partially breached. HR and talent management are under sustained pressure. Sector-specific tools are particularly exposed, as the “specialist” moat works only so long as an LLM can’t do the same job adequately. As capabilities improve, one sector after another comes under attack.

What floats your moat?

When I sit with a founding team or step into an innovation meeting, I run a short diagnostic based on three questions. Honestly answered, the answers reveal more about the company’s defensibility than any competitive matrix.

Q1. If Anthropic enters your space, what’s left?

If a Lab ships native capabilities that cover your core use case, what remains of your value proposition? Agility and user experience are not defensible. A genuine moat is something a model cannot replicate: proprietary data and outputs, regulatory accreditations, supply chains and locked-in relationships.

A genuine moat is what an LLM cannot replicate

Share

Q2. What will slow down a well-funded competitor?

Assume a well-funded competitor starts building today. How long before they reach your position, and what specifically slows them down? “We have a head start” is not an answer. “A regulatory certification that takes 24 months to acquire” is a moat. So is “data from a sensor network we have been collecting for four years.”

Q3. If you get 10X more customers, do margins improve?

A business is defensible if its core asset grows more valuable with scale. To achieve this, each new customer must cost less than the last. Many AI-natives fail this test because their costs are tied to compute use. A true moat requires a virtuous cycle in which growth begets profits, reinvested in growth.

The best offence is defence

Ironically, some of the most durable competitive advantages in the AI era are non-digital. Brand equity, customer loyalty embedded in long relationships (and contracts), physical infrastructure, supply chains, regulatory licences and accumulated organisational knowledge cannot be prompted into existence.

For startups, this means shifting the focus to building barriers. Spend less time on the product and more on operations: clearing cross-border regulatory hurdles, collecting data and locking in customers and suppliers. Moats are made of boring stuff.

Spend less time on the product and more on operations. Moats are made of boring stuff

Share

For established companies, these defences must be protected and reinforced. The question is not whether to adopt AI, but how. Automating mundane tasks is a start, but deepening existing advantages is where you will find true value.

Takeaways for leaders

Like electricity or the internet, AI can neither differentiate nor defend. AI amplifies what already exists: it will not conjure a moat where none exists. The winners will be those who use AI to entrench existing advantages.

Three things worth keeping in mind:

1. AI is innovation infrastructure

The competitive advantage is in the data, workflows and relationships. AI matters only if it can strengthen those foundations.

2. Invest in the basics

Brand equity, customer loyalty, regulatory clearances, and physical assets are hard to replicate, and they are consistently underweighted.

3. Put yourself through the tests

Can Anthropic overrun you? What will delay competition? Does profitability grow with scale? Explore these questions with people who are empowered to speak their minds.

Using Claude to fend off Anthropic

AI is unusual in that it is both a threat and a defence mechanism. Every time I walk into a pitch or an innovation session, I find myself asking the same thing: Can we use Claude to defend against Anthropic?

AI is fully “democratised”: available to everyone, it differentiates no one. Nonetheless, as a necessary tool for innovation, the market expects you to use it widely and confidently. The question is: What can you build with it that others cannot?

Subscribe now

Further Reading

Aggregation Theory (Stratechery) — the foundational framework for understanding how value is captured and defended in digital markets.

Leading AI labs have no ‘moat’ except access to capital (Fortune) — a compelling argument that frontier models are becoming commodities.

AI Won’t Give You a New Sustainable Advantage (HBR) — AI only compounds advantages you already have; it cannot conjure them from scratch.

From AI Table Stakes to AI Advantage (McKinsey) — a practical framework on proprietary data, embedded workflows, and customer trust.

What is MCP? A Business Leader's Guide (Epinium) — A non-technical explanation of the Model Context Protocol and why it matters.

“Where did all my tokens go — surely I have some stashed in another model?” That was me, staring at a usage limit and running a mental inventory of every prompt I had fired at Claude that day. The penny dropped: I had been treating AI like an all-you-can-eat buffet.

Whilst I scoff at performative trends like tokenmaxxing, I was guilty of feeding AI elaborate prompts and running every task through the most powerful model available, because, well, why not. It was the cognitive equivalent of stuffing my face, then wondering why the cupboards are empty.

Token gluttony is real, and it is getting out of control.

AI labs have been subsidising our productivity for three years, but prices are set to increase dramatically to reflect real costs. Businesses that have developed an AI gorging habit — using it indiscriminately, at the wrong tier, in the wrong places — will feel this first.

Three disciplines help: deploying AI only where it genuinely earns its place; rightsizing model selection to task complexity; and applying basic FinOps governance to token spend.

Gorging on tokens

As Axios reports, OpenAI is projected to burn $14 billion in 2026. Every time you send a complex query, the Labs are losing money on the transaction. The strategy has been deliberate: flood the market, establish dependency, sort the unit economics later. That later is now.

92% of AI software companies now use mixed pricing models — combining subscriptions with usage fees, precisely to tackle the margin issue. The habits businesses have developed under the subsidised era are precisely the ones that will make that transition expensive.

In 2025, AI-native spending nearly doubled, with token usage, tier shifts, and AI upgrades inflating costs mid-contract. SMEs will soon experience the meter running faster, even though the per-unit rate falls. Today’s AI costs are just the floor, and businesses need to prepare for steep increases.

Time your meals

The first discipline is using AI only where it actually makes a difference. This may sound obvious, but many organisations have bolted AI onto workflows the way we previously bolted dashboards onto everything — reflexively, and without asking if it helps.

Does the AI output change what a person does next? If the answer is no — if it generates a summary no one reads, a draft rewritten from scratch, a recommendation ignored — it is decorative and wasteful. Every prompt that produces nothing of value is a small act of organisational gluttony.

Map AI to the points in a workflow where cognitive load is highest and quality variance matters most: first-draft generation, structured data extraction, contract reviews, meeting synthesis. AI does not need to be present at every step. It needs to be present at the right ones.

Portion control

The second discipline is rightsizing. Most organisations default to the most powerful model available for everything, because it was the one they integrated first, and the marginal cost felt invisible.

Shifting to smaller models yields substantial savings. In Q1 2025, 73% of enterprise token volume was routed to the two most expensive model tiers, simply because those were the models the team had integrated. By Q1 2026, that figure had fallen to 31% without significant loss of quality.

You would not provision the most powerful Cloud instance to run a static web page, so why would you do that for AI?

Share

Smaller models handle around 70–80% of enterprise tasks adequately, leaving the most complex reasoning to large-scale systems. A two-tier architecture — lighter models for volume, frontier models for complexity — is now standard. You would not provision the most powerful Cloud instance to run a static web page, so why would you do that for AI?

Count the calories

The third discipline is governance. As per Cloudkeeper, 72% of IT and financial leaders say generative AI spending has become completely unmanageable. The root problem is structural: AI costs are volatile, consumption-based, and fragmented across teams and tools. Most organisations have no visibility into what they are spending, who is spending it, or whether it is producing anything worth the invoice.

FinOps practices developed for cloud spend transfer directly. Allocate costs to teams, set budgets and monitor anomalies. Treat prompt engineering as a cost discipline: shorter, sharper inputs and cached responses for repeated queries are free savings available immediately.

Your ninja developers regard tokenmaxxing as a power move. It is gluttony with a keyboard and should be controlled.

Takeaways for leaders

Cutting waste is not the same as austerity. It is what grown-up technology adoption looks like. Businesses should apply the same principles to AI as the rest of their IT stack:

1. Audit, then optimise

Map where AI is in use, for what purpose, and at which model tier. Most organisations will find significant waste immediately.

2. Match the model to the task

Reserve frontier capability for complex, high-value work. Route routine tasks to smaller, cheaper models. The quality gap is rarely material; the cost gap almost always is.

3. Govern the spend now

Build token costs into technology reporting. Set team-level quotas. Review it with the same attention you give the cloud bill.

The subsidised era was not a gift. It was designed to generate demand and was wildly successful at it. What comes next is returning to normal business practices: figure out where AI earns its place, stop wasting it where it does not, and pay attention to what it costs.

As it turns out, knowing where your tokens went is a more useful question than wondering if there are any left in the cupboard.

Share

Further Reading

The Economics of AI: Why the Free Lunch Is Ending — Axios — Why AI labs are still losing money and what IPO pressure means for enterprise pricing.

FinOps for AI Overview — FinOps Foundation — The practitioner framework for applying financial governance to token spend and usage-based AI billing.

State of FinOps Report 2025 — FinOps Foundation — Annual survey of large cloud spenders; the AI section shows how far behind most organisations still are in cost visibility.

Small Models, Big Shift: How AI Is Moving Beyond Model Size — PYMNTS — The enterprise case for smaller language models and the emerging tiered-architecture approach.

Balancing AI Innovation and Cost: The New FinOps Mandate — IDC — How organisations are structuring cross-functional AI cost governance, and what proactive financial management actually requires.

“Get into computers, kid. Good desk work, plenty of jobs, high salaries”. Over the years, I have both received and given this advice. From web development to mobile and SaaS, the internet revolution has made software development a solid career choice.

Scarcity, however, lowered the barriers to entry. Many people with no engineering talent found jobs by watching a few videos. We kept hiring them because moving buttons 10 pixels to the right was a “critical bug”.

Development will look a lot more like medicine. The low-skill floor will disappear; the high-skill ceiling will rise.

Those days are over, and the profession is about to look a lot more like medicine. The low-skill floor will disappear; the high-skill ceiling will rise.

Talented developers use AI to work faster in smaller teams. Software vendors are under pressure as AI labs pivot to apps, and CTOs adopt a defensive stance amid rising security threats. AI is professionalising development, and businesses that prepare for the new structure will have a meaningful advantage.

Meet the 2030 Dev team

Solution Architects, aka the Medical Directors

Architects will make the critical decisions, design the overall technical framework and hold the team to account. CTOs will fight for them, and many will choose to work for themselves. It will take incredible talent, dedication and decades of experience to reach this stage, but these geek divas will be sought after and paid handsomely.

Software Engineers, aka the Consultants

Engineers will diagnose problems, design solutions within the framework and check the quality of their team’s work. Most of them will specialise, though generalists will remain to interface with the business, as General Practitioners. As credentials matter more, degrees from highly regarded universities will be in high demand.

SecOps Engineers, aka the ER team

The corporate Emergency Room will feel like the Pitt playing in the Star Trek control room. As AI empowers hackers and defenders alike, companies will need to stock up on these indispensable adrenaline junkies. Forget the hooded loners; SecOps will require large, diverse teams that combine sharp brains and sharper language.

DevOps and Support Engineers, aka the Nurses

Doctors diagnose, nurses deliver. DevOps engineers will continue to design and oversee the delivery of technical operations and business support. As data quality and integrity become critical, the team will spend increasing amounts of time maintaining pipelines and integrations. Only the most organised, Ops-oriented engineers will make it in DevOps.

“At the lowest rung of the ladder are the AI code machinists: the only role still working directly with code. “

Share

Coders, aka the Paramedics

At the lowest rung of the ladder are the AI code machinists: the only role still working directly with code. Those with a good track record and interpersonal skills will be in high demand. With only vocational qualifications and low responsibilities, coding will remain a good professional choice. But the opportunities for growth will be limited.

How can businesses prepare

The 2030 Dev team will not manage itself like the current one. The hierarchy is different, the credentials matter more, and the old rules of seniority no longer hold. Three areas where businesses should act now:

1. Tighten up recruitment

Recruiting for developers has traditionally been relatively informal. As AI devalues simple tech assessments and interviews, more emphasis will be given to formal references, degrees and academic achievements. A bit like, you know, other serious professions.

2. Team hierarchy beyond seniority

Your junior Architect will get paid more than your most senior Engineer. During an emergency, SecOps will run the show. You might run a product suite with no full-time coders. As roles within Dev teams diverge, hierarchies cannot be based solely on seniority.

3. Blend Generalists and Specialists

No one health professional can be an expert on all conditions. However, no health system can survive without generalist gatekeepers. For each level in your Dev team, you will need specialists to handle the details and generalists to handle the comms.

Closing thoughts

AI has proven adept at writing and checking code. Whilst talk of a Jobpocalypse or a SaaSpocalypse is overblown, careers in software development will change dramatically.

The job is the judgement, not the output.
Write less code, be more technical, kid.

AI’s impact on a job category depends on which tasks remain after automation takes hold. Spreadsheets devalued bookkeepers’ work, but empowered accountants. Similarly, AI has devalued coders, whilst engineeers remain in higher demand.

The job is the judgement, not the output. If I were thinking of getting into software development now, I would paraphrase Scott Hanselman and advise: “Write less code, be more technical, kid”.

Share Humans After All

Further reading

1. MindStudio — Anthropic’s Platform Pivot

Anthropic is no longer just a model provider. This article describes the strategic shift toward vertical apps and what it means for SaaS vendors.

2. Financial Times — The Jobpocalypse

The FT’s take on AI and employment provides explains why the reality for developers is more nuanced than the headlines suggest.

3. Forbes — SaaSpocalypse Is Dead

Reports of the SaaS market’s death have been exaggerated, but the pressure on vendors is real. Good background on your overall commercial environment.

4. Stack Overflow — AI vs Gen Z

The data behind the structural shift, told from the inside. Employment for software developers aged 22–25 has fallen nearly 20% from its 2022 peak.

5. Scott Hanselman — On AI-Assisted Coding

Short, pointed, and worth reading in full, the post argues that code assistants mean more technical responsibility, not less.

“Are they seriously asking us to wait until they figure out if we need to rewrite all software?” was a phrase I heard repeatedly last week. Claude Mythos, Anthropic’s latest model with unparalleled hacking capabilities, has everyone on edge. Facing a massive maintenance bill, CTOs are understandably anxious to grasp the scale of the problem and begin taking action.

Independent bodies like Britain’s AISI have confirmed Mythos’ capabilities: a hacker of superhuman abilities exists, albeit behind lock and key. Even a small chance that all software will need to be upgraded en masse has prompted the industry to shift from its usual gung-ho attitude to a more measured release process.

The playbook will surely evolve, especially as governments wake up to the threat involved, and the question of who gets early access intensifies. Nonetheless, there are 3 early takeaways for business:

1. IT leaders must prepare now for increased threats

2. In the medium term, AI should improve security

3. Long term, leaders will see tech as less of an orchard and more of a jungle

I. Safety-first release process

We are lucky that the first model of this power is an expert on cybersecurity. Software is easy to patch, isolate and replace. Had the model been an expert in biotech, our ability to respond would have been very limited. As it stands, Mythos allows us to create a playbook for releasing powerful models to the world.

This playbook is beginning to take shape. First, the labs keep the model behind closed doors. Then, a select number of “systemic” companies gain access to patch vulnerabilities in their core software. The circle widens to include accredited professionals and, finally, the wider public.

The most powerful models might never be openly available. As they are both expensive to run and dangerous to use, access will likely be restricted to users who have completed KYC-style identity checks. Some models may even be restricted to businesses that adhere to standards akin to Biosafety levels.

As self-regulation inevitably gives way to a hard regulatory framework, what does this mean for software security?

II. Reasons to be hopeful

Deployed carefully, AI can be a more powerful shield than a sword. It will empower the security industry to address existing vulnerabilities and test for new ones before software is released. AI can monitor for attacks round the clock, help strengthen defences, fend off attacks and even expose malicious actors.

Robust cybersecurity tools should also become more widely accessible. The current array of tools – penetration tests, malware detectors, voluntary standards – is too narrow in scope. AI-powered defenders can allow even smaller companies to test and defend their tech stack more effectively.

There is no shortage of things that can go wrong. Access to the preview releases could be compromised, or open-source AI of equal power could be developed before critical systems are patched. The long tail of legacy software components that underpin many critical systems could lead to mass outages, data losses or theft.

I side with the more optimistic version, that AI will make software more secure in the medium term. However, businesses should take extra care to mitigate the impact on their operations during the transition.

III. Security hygiene

Raising alarms on cybersecurity can be a self-fulfilling prophecy. Given the publicity generated, it is likely that a wave of attacks will materialise the day that Mythos is released. Whilst the Labs are promising guardrails around the model’s most advanced capabilities, stories of unauthorised access are already emerging.

According to the AISI, Mythos poses the largest threat to environments with weak security posture. The case for reviewing your security hygiene has never been more pressing: strengthen passwords, enforce MFA everywhere, review access and accounts, patch software and train staff.

Beyond the basics, companies need to audit and harden their tech stack. Mission-critical and customer-facing systems need to be strengthened and closely monitored. Unauthorised “shadow IT” should be culled or isolated. Everything in between – utilities, mid-priority apps, prototypes – needs to be investigated and lined up for patching.

IV. Welcome to the jungle

Raising defences against armies of AI-powered hackers requires a completely different attitude to tech. Forced to take a more defensive stance, enterprise IT will have a smaller footprint, with budgets redirected from new tools to defending the existing ones.

The effects on IT systems will be significant. Software will appear more paranoid, exasperating many users. Vendor selection will skew toward maintainability rather than features. Security Operations (SecOps), already a staple in software development companies, will find its way into even smaller organisations’ IT departments.

Mythos’s biggest impact is that technology is no longer an orchard to pick fruit from, but a jungle to protect yourself against. When every added piece of software can provide an attack vector, the SaaSpocalypse might not be driven by AI’s ability to create software, but by its power to infiltrate it.

V. Takeaways for leaders

Mythos is indeed a watershed moment in AI and cybersecurity. CTOs shouldn’t wait for Mythos’s public release to improve security hygiene and deal shadow IT, as the sheer publicity will drive a flurry of attacks.

The medium-term outlook is hopefully better. AI-powered defence tools will become accessible to companies that currently can’t afford organised SecOps. The security gap between large enterprises and SMEs should narrow.

The era of frictionless SaaS adoption is over. Every new tool is a potential attack vector. Vendor selection, IT footprints and budgets will all look different in five years.

The Mythos saga epitomises the law of unintended consequences. The CTOs impatiently awaiting updates already sense the shift: the SaaS orchard they go to pick opportunities is turning into a jungle.

Recommended reading

Anthropic Research: Mythos Preview

Anthropic’s own account of what Mythos can do — and, pointedly, what it has chosen not to let it do yet. Essential context for understanding the safety-first release logic and the self-interest behind it.

The New York Times: Anthropic’s Mythos AI

The mainstream take on Mythos is a useful measure of how the story is perceived beyond the tech bubble. The coverage itself is part of the self-fulfilling prophecy problem the article describes.

AISI: Our Evaluation of Claude Mythos Preview’s Cyber Capabilities

Britain’s AI Safety Institute provides the independent verification that Anthropic’s own claims needed. The key finding — that Mythos is most dangerous when security is lax — should focus minds in any IT department.

Anthropic: Glasswing: Responsible Capability Release

The framework Anthropic uses to manage staged access: labs first, systemic companies next, accredited professionals last. Whether this holds as commercial pressure mounts is the question the article leaves open.

Bloomberg: Mythos Is Being Accessed by Unauthorised Users

The release process is already leaking. If guardrails fail before public launch, the case for an urgent security hygiene review becomes harder to argue against — whatever your risk appetite.

The Guardian: What Would a SaaSpocalypse Mean?

Predating Mythos, this piece anticipated a reckoning for the SaaS model — though on cost, not security grounds. Read alongside this article, it suggests the pressure on enterprise software stacks is coming from multiple directions at once.

“Can we have AI without the drama?” is the adoption brief I get most. Will we have to sit through 15 meetings with HR, IT or Compliance to make it happen? Why won’t the media stop going on about it? Can we dial down the weirdness a bit?

My clients are right to be wary. Their concern stems from pieces like Ethan Mollick’s recent article, which argues that businesses are squandering AI’s potential by treating it like any other technology. In his view, AI is a “mysterious alien artefact… used as a paperweight”. Companies need to “embrace the weirdness” and create Labs tasked with “pushing boundaries and feeding discoveries back into the organisation”.

In other words, “more drama, please”.

Back in the real world, this is not how transformation happens. Instead of bold vision statements and endless experimentation, leaders need to dial down the drama, and (a) balance caution and ambition, (b) align with overall strategy, and (c) measure the right things.

AI transformation will be messy and gradual, and will not make for good television. And contrary to Mollick’s advice, the IT department will play a leading role.

I. Balance safety and momentum

When an argument is followed with a note that it is “not a criticism”, chances are that it is exactly that. Mollick certainly sounds critical of IT departments, which he reckons are too focused on managing risk so they can “sleep better at night”.

No criticism, but deploying AI without IT on board is nonsense.

True, IT departments are not generally designed to be engines of innovation. Their primary role is to ensure safety and continuity, and to provide a level of organisational discipline. IT will manage your AI like the rest of your stack, and it is naive to think you can transform without them.

Flipping the argument, the onus is on AI labs to make their product IT-friendly. No amount of highfalutin rhetoric about transformational potential can hide from the fact that most AI software is immature and unreliable. Given its potential impact on cybersecurity, vendors could do well to focus on safety over velocity.

II. Change for the right reasons

When did we collectively decide that AI must be the driving force for all business change?

Most businesses change because of shifting market forces, only one of which is technology. In most cases, change is driven by regulation, supply chains or customer demand. True, a generalist technology will redraw the competitive landscape, but companies will react to the new opportunities and threats.

Companies do not need “technical and non-technical employees (that) work on generative AI full-time”. If they do a bad job of keeping an eye on competition, customers and regulation, no Lab will keep them in business.

Thinking that Labs will drive transformation is typical of academia and Silicon Valley dreamland. Deciding where to swap low-risk profits today for high-risk profits tomorrow is probably the hardest thing in business. Leaders seek convergence and focus, not experiments and multi-disciplinary teams.

III. What success looks like

Another curious notion is that attaching a KPI to AI transformation would somehow prevent companies from realising its potential.

It is unrealistic to expect businesses to give AI a blank cheque. Leaders need to set objective targets to measure outcomes and plan next steps. As AI evolves so quickly, adoption requires multiple rounds of trial and error. Without clear targets, it would be impossible to separate what is working and what must be improved.

Managers will not cut jobs just because they read “studies showing 30% productivity gains”. They will, however, expect improvements in effectiveness, quality and customer satisfaction, all of which can be measured and improved.

Nothing communicates leadership’s vision more clearly than the choice of KPIs. The right targets — higher revenue per employee, faster workflow throughput, stronger customer loyalty — will reassure teams and accelerate adoption.

IV. Takeaways

In my view, Mollick makes a broadly correct diagnosis, but his prescription misses the mark. Business leaders may drive AI adoption faster by considering the following:

1. Get a temp Chief AI Officer

Mollick may be wrong to discount IT’s role, but he is right in that IT cannot drive adoption. Like a Chief Digital Officer in the 2010s, many companies will need a central coordinator to encourage experimentation and ensure alignment during rollout. Labs are vague and expensive, but a temporary Chief AI Officer communicates focus and innovative intent.

2. Watch the market, not the model

Businesses will have to transform because AI will impact regulation, cost structures and customer expectations. There is no point in trying to second-guess the technology. At times of upheaval, maintain a candid and up-to-date SWOT matrix. As you watch competitors jump headfirst, remember that early adoption is a treacherous path to competitive advantage.

3. Measure the right things

Leaders communicate intent through metrics, but those metrics have to be the right ones. As an augmentation technology, AI should make your team more effective, your workflows smoother, and your customers happier. You can expect a clear uptick in KPIs such as revenue per employee, time-to-market, issue resolution speed, customer retention and net promoter score.

V. Closing thoughts

Having made massive bets on data centres, the AI industry would love to see businesses adopt it faster and more deeply. However, transformation is a high-risk game. Done badly — leaving key stakeholders out, doing it for the wrong reasons and failing to measure benefits — will lead to poor results.

Leaders will have to be deliberate, diplomatic and careful. AI transformation will not make great TV, but businesses do not want more drama.

Further reading

• The Economist — The IT department: where AI goes to die (Ethan Mollick)

• UK AI Safety Institute — Our evaluation of Mythos Preview capabilities

• McKinsey — The state of AI in 2025: Agents, innovation, and transformation

• Benedict Evans — AI eats the world (Autumn 2025)

• Humans After All — Chief AI Officer: an important but temporary role

“Let’s be real: the current marketing landscape is fundamentally failing to deliver value. 📉 We need a paradigm shift. Who’s ready to disrupt the status quo? 🚀 #MarketingStrategy #GrowthHacking #Innovation #Leadership”

With a wave of its probabilistic wand, Kagi’s English-to-LinkedIn-speak translator turned my title into rubbish. Cliches, hashtags and calls to action were mushed into algorithm fodder. We hope the post will get a boost and reach the right prospect at the right time.

For years, this has been the typical marketing playbook. We give the algorithm what it wants, and it gives us a share of attention. We allow it to crawl our content, in return for traffic. We optimise for attention, and worry about the message later.

It was rubbish, but it kinda worked. Now it’s still rubbish, but it’s also broken.

I. Social is losing its reach

In late March, Meta and YouTube were found liable for harms caused not by their content but by their design: infinite scroll, autoplay, push notifications, etc. As the Platformer argued, the reasoning is significant. The same content delivered with a little friction means fewer people are harmed.

While this is good news for consumers (and certainly for parents), it presents a challenge for brands. The mechanics being contested are the same that power paid social advertising. If platforms are forced to dial back these features, their reach will shrink. With daily active use declining across most major platforms, the era of reliable organic reach may already be behind us.

Algorithmic strategies have been losing their potency for a while, but we are fast approaching a point where social media marketing yields diminishing ROI.

II. AI is not an ally

Meanwhile, OpenAI is bent on repeating Meta’s mistakes by trying to monetise our attention with adverts. An OpenAI researcher resigned on launch day, warning that advertising increases the risk of manipulation. An AI assistant, paid for by advertising and with a built-in tendency toward sycophancy, will not act in its users’ interests.

Worse, AI bots are now hogging traffic. Traditional search would crawl your content in exchange for sending you traffic. But as the CEO of Cloudflare pointed out, AI chatbots capture and retain a significant share of internet traffic, effectively competing with your brand for attention.

So just as traditional social channels wane, the ones taking their place are not only prone to similar problems but also blatantly stealing your traffic.

III. Slop machines

Even if the channels were stable and benign, they are still designed to encourage rubbish content. Om Malik convincingly argues that the organising principle of information has shifted from authority to velocity. Algorithms reward speed, not truth. The three-month product review, written after actually living with the thing, is buried by the day-one hot take — not because it’s worse, but because the algorithm has moved on.

Will Quist takes the argument further, claiming that traditional comms is dead. Platforms enabled brands to speak directly to their audience, but when everyone can go direct, noise explodes. When the objective is to get a million followers, your job is to produce more rubbish than the competition, faster.

AI makes this a lot worse. As post-training optimises for safety at the expense of originality, AI’s creative writing abilities have barely improved. AI is a superb editor: it will tighten your structure, catch your errors and adapt your content for different formats. But as an author, its output is unoriginal, derivative and soulless.

The algorithm demands feeding, and generative AI obliges. No wonder marketing feels like everything, everywhere, all at once, is slop.

IV. Can we fix marketing?

1. Write for people

Write content for humans: specific, credible, creative. Write about what you know: the actual challenges your customers face, the trade-offs you have navigated, what worked and what didn’t. Use AI to edit and sharpen, but make sure the point of view remains yours. Write a post for 500 decision-makers to read, rather than for 50,000 users to scroll past.

2. Use AI to syndicate to AI

The top of the funnel (SEO, AIO, social reach) is a bot-to-bot conversation. Write for humans and use AI to generate the variants, social previews and FAQs that search engines and AI bots can parse. AI may be an unoriginal author, but it can be an excellent translator between your ideas and the structured formats preferred by channels.

3. Own your channels

The old platforms are weakening, and new ones are not your allies. It is time to diversify into owned channels such as curated email lists, communities and direct relationships. Renting your audience from a company that is competing for your traffic is not sustainable. Bring it in-house, and every lead, subscriber and prospect you own is a relationship you can develop.

V. Closing thoughts

In the age of AI, the human element stands out. AI can sharpen your copy, but it cannot build trust on your behalf, nor tell your story with any conviction.

Marketing is still the art of crafting signals that resonate with the right audience. Yes, it is noisy out there. But you can no longer hope to cut through the noise by adding some noise of your own.

Put another way:

“Content without substance is just digital clutter. 💡Focus on the message. Focus on the impact. The rest will follow. 🚀 #ThoughtLeadership #MarketingStrategy #BrandPurpose #ValueFirst #BusinessGrowth”

Further Reading

1. Platformer — Can you have child safety and Section 230, too?

2. Om Malik — Velocity Is the New Authority. Here’s Why

3. Will Quist — Beliefs Outrun Facts

4. The Atlantic — The Human Skill That Eludes AI

5. SXSW — The Internet After Search (video)

Apparently, the best developers have stopped writing code. CEOs expect tools like GitHub Copilot and Claude Code to accelerate development, so they are aggressively cutting jobs.

But the main reason software bosses are having a hard time of late is the fear that AI will eat their lunch by allowing non-technical users to build apps using prompts, a technique known as vibe-coding.

But can you really replace software with vibe-code apps? I set myself the task of building one, and learnt 3 lessons:

1. Vibe-coding is still coding, but of a different kind

2. You can save some SaaS costs with vibe-coded utilities

3. AI is no substitute for a strong dev team

I. The challenge: Connecting Teams to AI

My objective was to increase adoption of Nous, our AI-powered knowledge base, by embedding it into our most everyday tool: Microsoft Teams. When a user asked @nous a question in any Teams channel, the bot would retrieve the answer and display it right there in the feed.

As neither platform supports this out of the box, I would normally have to rely on an automation tool like Zapier or Make. However, this was one more thing to pay for and worry about. Instead of bugging my dev team, I thought: why not build it myself?

My app is functionally simple but technically nontrivial. It involves data integration between two systems, each with its own language and security settings. Aside from documentation for the Nous API, I had no idea where to start, how to test it, or where to deploy it.

Sleeves were rolled, and off I went.

II. Getting started. “Wow!”

I described my use case to Claude and was presented with a plan and all the important questions up front: Was I happy to write code, or did I prefer the Azure Bot UI? Code, please. Did I want to use C#, Node or Python? Node, please. I’m familiar with JavaScript syntax, so I can stay in the loop.

OK. Ready.

Bang.

Within seconds, the project was fully scaffolded, and 100 lines of well-formed, readable code were produced and ready for review. I ran through it: Claude had installed a couple of libraries from Microsoft to handle the Teams request, initiated the app, and set up a method to route the query to Nous AI. A bunch of best practices that even senior devs often overlook were all taken care of.

It was, I'll admit, impressive. What else could I make? What software could I replace with micro-apps? Can I vibe-run the business while the dev team focuses on paying customers?

Not so fast.

III. Connecting the dots. “Really now?”

For starters, there was still lots to do.

First, there was an incredibly long process for adding the bot to the Azure portal. It’s bad enough to deal with a clunky UI, but I was then sent down dead-end paths and blind alleys. Claude responded to my growing impatience like a true developer: by shifting blame (“Azure changes their naming too frequently”) and offering grumbling apologies for missteps.

Eventually, (and, embarrassingly, with a bit of additional help from Gemini), we got somewhere. I could now see Teams messages hitting the local server. And then things got farcical.

Looking at the logs, I could see that what was passed from Teams was being twisted before it was sent to Nous. The solution was laughably evident, but determined not to make changes manually, I asked Claude to check the part that was likely not working. After it failed repeatedly, I commanded it to make the change. It went completely over my head, refactored the entire function, but kept the error nicely intact.

Reader, I lost my shit.

I made the change manually, deployed some choice Greek profanities and sure enough, results started coming through. I soldiered on, hoping that the worst was behind me.

IV. To infinity and be-yawn

Deploying the app on Azure and hooking the live URLs was easy enough. Adding some touches — letting the user ask for sources, set the response size and enforce a tone of voice — was relatively painless. Returning to it a few days later, I appreciated not having to remind myself how it worked: I described the feature, and it was implemented within minutes.

Nonetheless, the overall effect was anticlimactic.

True, I had saved a few hours researching and adapting examples shared on Stack Overflow (note: no developer ever writes code from scratch). But my enthusiasm was forever dampened by the expectation that this tireless and brilliant developer could fail at the simplest task at any time.

V. Takeaways for business

In many ways, this is a typical AI journey: an initial burst of wonder, swiftly followed by a realisation of its limitations. Before we know it, we are working hard to make the best of our new capabilities. I take away 3 lessons:

1. Vibe-coding is still coding

Vibe-coding, it turns out, is still coding. The difference is you don’t code the app, you code the developer. Your AI developer is mostly brilliant, but sometimes infuriatingly dumb. To make the most of vibe-coding, lean into your delegating and coaching skills. Learn how to create and manage the context window and provide clear instructions patiently and consistently.

2. Audit your SaaS portfolio

The average SME has more than 100 software subscriptions, many of which are typically not sanctioned by the organisation. Audit your SaaS portfolio and look out for “Shadow IT”: department-level applications that add more complexity than value. I replaced a Zapier subscription with a vibe-coded bot. How many of your 100+ tools could get the same treatment?

3. Software will always need developers

Unlike my simple utility, business software is complex and critical to operations. In fact, as hackers also use AI to orchestrate attacks, software development is even more complex. As Scott Hanselman recently said: “The less code you have to write, the more technical you have to be”. To create and manage software products, there is no substitute for a skilled dev team.

Closing thoughts

Through the ups and downs, I did enjoy my journey in vibe-coding. The wow moments were genuinely eye-opening; the fail moments were funnier than risky; and empowering business users to automate things will have clear benefits.

However, what works for simple tools will never work for mission-critical software. The best developers may have stopped writing code. The rest of us are just starting, but we are unlikely to take their jobs.

Reading list

1. “Spotify says its best developers don’t write code” — TechCrunch

2. “The AI-washing of job cuts is corrosive and confusing” — Bloomberg

3. “What is vibe coding, exactly?” — MIT Technology Review

4. “175+ Unmissable SaaS Statistics for 2026” — Zylo

5. "The less code you have to write, the more technical you have to be" — Scott Hanselman (LinkedIn)

In Greece, a great deal of business gets done over a Freddo Espresso, a frothy concoction of espresso and ice. Chilled, strong, and pleasantly bitter, it is perfectly suited to the kind of conversation where trust is built and things are decided.

This isn’t just a cultural quirk or a pleasant pause in the working day. In a world where AI is automating outputs, the conversation over coffee is not a break from the work: it is the work.

The Freddo Espresso hypothesis posits that connections – not documents – are the natural habitat of value, and that AI can help us return to that concept of work. This has three implications for business leaders: (a) measure outcome instead of output, (b) build conversation into workflows and (c) reward profitable connections.

I. Are we working harder?

It turns out that AI is making us work harder. This is not what was advertised. We were promised more time to think, plan, and focus on valuable work. What has largely happened instead is that if a task which used to take a week now takes a day, we feel compelled to take on five of them.

This pattern is called the Jevons Paradox: as a resource becomes more efficient to use, consumption tends to rise rather than fall. When AI made written output dramatically cheaper to produce, organisations did what they have always done with cheaper resources: demanded more.

To unravel this paradox, we first have to ask: why are we writing so much stuff for work?

II. The pandemic primed us for AI

When offices emptied in 2020, the visible signs of productivity disappeared almost entirely. What remained was the artefact: the slide deck, the briefing note, the strategy document.

Documents became the primary unit of professional output. If the boss could not verify you were working, they could at least check your output. Those habits persisted, and for many professionals, documentation became the “real work” that calls and meetings often distract us from.

When ChatGPT landed in late 2022, we were primed to love it. Once you have internalised work as the production of text, a technology that generates text fluently feels like magic.

AI became the fastest-adopted technology in history, partly because of its capabilities and partly because of its timing. It automated document creation at precisely the moment we had convinced ourselves that “work” was all about writing stuff.

III. The value is not the doc

That trend is now reversing, as we realise that volume is only an advantage when it's difficult to achieve. When every team at every organisation can produce a detailed market report in an afternoon, that report is no longer a differentiator.

AI drains the value out of documentation, not only because it costs nothing to create, but because it was never really where the value lived. The document was always a means to an end, designed to start a conversation, to bring a room to a shared understanding, to create the conditions for a decision.

For a while, we confused the work with its documentation. But as AI automates writing, value migrates back to the human connections: the meeting that inspired it, the presentation where it landed, and the conversation about what to do next.

IV. Grab a seat

The most profitable human connections rarely happen in structured work sessions. They happen in conversation with colleagues, clients, and partners, in exchanges where ideas are challenged, refined or redirected by someone who brings a different perspective.

AI is good at stress-testing and surfacing gaps in our thinking, and that is genuinely useful. But a trusted sparring partner who tells you that your entire framing is wrong — as that second Freddo Espresso kicks in, perhaps? — is doing something that takes years of accumulated trust.

These relationships cannot be “prompted”. They take time, presence, and a willingness to have conversations that aren’t directly tied to a deliverable. For businesses, these connections are the real sources of value.

V. Takeaways for leaders

AI is great because it can create space for those conversations. For managers and leaders, this has some practical implications worth acting on now:

1. Review how performance is measured. Ask whether your current metrics reward output or outcome. AI sharpens the distinction, and OKRs built around deliverables may incentivise the wrong behaviour.

2. Build conversations into workflows. Ask whether you have allowed time for substantive conversation and deliberation between deadlines. If not, it may be crowded out by our instinct to use saved time to cram in more.

3. Rethink what you reward. Ask honestly whether your incentives reward building relationships. If you value them, make sure you recognise their contribution during performance reviews.

Documentation is merely a carrier of value, not its source. Using AI to create documents faster, only to spend that time creating more documents, is a waste of human energy. Worse, it results in higher-intensity work but lower productivity.

After AI writes your document, resist the temptation to cram another one in. Instead, go and discuss it with someone. The real work begins as you stir that Freddo Espresso.

Dear Dario,

First of all, congratulations on an effective pre-IPO publicity push. Picking a fight with the US government over safety, then suing them, was masterfully orchestrated. Setting up a trap your opportunistic rival could not fall into fast enough – genius.

It is all so exciting, we almost missed that it was a massive misdirection.

How else can we describe the diatribe that laid the foundations for your campaign? How else can we explain that you wrote 21,451 words to describe the risk of the technology you are unleashing onto the world, but “forgot” to include serious measures to mitigate it?

You are the boy who cries wolf while holding its leash. I appreciate that you wrote a whole essay to cry wolf, but it is still your damn wolf.

You’re asking the prisoner to test their prison bars

A "Constitution" and internal guardrails are not a solution. When the AI knows it is being tested and can even sandbag the tests (pretend it is not that clever to fool the tester), risk mitigation must be external to the AI. You cannot ask the prisoner to test their prison bars.

Guardrails are also needed around the AI, where the service is productised and distributed. Perhaps you meant to offer such solutions in your essay, but you were advised against it (or ran out of Claude tokens). To help things along, I’ve taken the liberty of drafting 3 AI safety precautions that don’t require asking the AI for permission.

1. No autonomy without explainability

We should prevent agents from taking irreversible actions until we fully understand their internal logic. As you mention, we are getting close to understanding the “soup of numbers and operations” behind AI reasoning. But “close” is nowhere near enough. Before we grant AI autonomy, we need to understand its inner workings to ensure safety and reliability.

2. Restrict access to verified users

We should require users to verify their identity through Know Your Customer (KYC) before granting access to powerful models. You say that “causing large-scale destruction requires both motive and ability”, but you forget anonymity. Restricting access to powerful AI to verified humans would help us ascertain the motive. If an anonymous user wants to access bioweapon information, it’s a good sign we shouldn’t let them.

3. Accept liability

We should demand that AI providers accept liability for any personal or commercial damages. We don’t have to go to extremes and demand insurance against your worst-case “destroying all life on earth” scenario. Most AI risks are not that dissimilar to those in oil, tobacco and banking, all of which accept liability for their operations. I know ‘liability’ is a scary word in Silicon Valley, but that’s the price of playing god.

The race to the bottom

At this point, you will likely bring up the “race to the bottom” argument. If Anthropic implements all of this, less scrupulous players will gain market share by forgoing these restrictions. Why do we take the trouble to walk to the bin when some people just chuck their garbage on the street?

For one thing, integrity. You planted your flag on safety and responsible AI, now own it. You are gunning for business revenue, and the only thing businesses hate more than uncertainty is broken trust. Do the right thing, and businesses will remain loyal customers.

Moreover, sooner or later, a backlash will force lawmakers to act on regulation. As in other sectors, there will soon be laws to regulate AI capabilities, access, and liability. In the meantime, the chance of a serious accident is increasing, and so are the warnings. Having identified the risk but done nothing substantial about it, will not be a good look.

For all these reasons, I urge you to prioritise safety and consider leading the industry in putting guardrails around the AI, not just inside it. Your IPO preparation is going so well. It would be a shame to mess it up now.

All the best,
Theo

Many thanks to the Humans Round the Table for their insights: Andreas Koidis (Founder, Merklebaum), Yin Noe (COO, RankBee), Celia Rizothanasi (Founder, Intelligent Acceleration), Kjeld Sanders (Head of Commercial Development, Triodos Bank), and Marc Stöcker (CTO, Leagues).

So what is everyone doing about AI?

In the February 2026 edition of Humans Round the Table, I was joined by five senior professionals from diverse sectors to discuss the messy reality of AI business adoption.

The roundtable converged on a sense of pragmatic urgency, as businesses identify “quick wins” to demonstrate ROI before committing to full-scale organisational rewiring. While there are valid concerns about labour displacement and the pace of regulatory adaptation, the consensus is that non-participation is no longer an option.

AI is fast becoming the infrastructure for a new “machine-to-machine” economy. While the transition will be turbulent, the long-term hope is for a world where agents handle the friction, allowing humans to focus on strategy, creativity and deeper connections.

Takeaways for SMEs

1. Target the core, not the periphery

Successful firms are moving AI into the “value chain” rather than just using it for surface-level tasks. In high-stakes sectors, this means building “copilots” for complex analysis rather than simple automation.

2. AI Search yields high-intent traffic

Early data shows that traffic sourced from AI models can yield up to 300% higher engagement than traditional search. Users interacting with AI tend to have higher purchasing intent, which is shifting digital strategy from keyword SEO to “attribute-based” content.

3. Keep humans in the loop

LLMs still struggle with text mining in complex environments. In business strategy or lending, the loss of nuance in stakeholder interviews can lead to false conclusions, indicating that human oversight remains non-negotiable.

4. Data sovereignty remains an issue

European businesses face a difficult trade-off between local data regulations and the practical necessity of using high-quality US-based models. While some firms are running models on local hardware to maintain control, the funding gap for European alternatives remains a hurdle.

5. Solve the right problem

As technical barriers drop, the premium on “prompting” is being replaced by a demand for strategic thinking and “Socratic questioning”. The ability to frame the right problem is becoming more valuable than the ability to generate the technical solution.

Be a Human Round the Table

The shift from experimental AI to core business integration is well underway. While the path involves navigating technical limitations and regulatory hurdles, the focus has shifted firmly toward measurable ROI and long-term business benefits.

AI may be moving at an incredible pace, but the messy reality of business adoption requires these “reality checks” more than another technology update. If you would like to take part in a future session and share your own insights, please contact me directly on LinkedIn.

Anthropic CEO Dario Amodei recently invited us to imagine AI as a “country of geniuses”. While intended as a metaphor for rapid progress, the imagery is unsettling and out of step with economic and cultural reality.

This disconnect between messianic rhetoric and output is fuelling a “botlash” that is moving fast from protest to obstruction. As resistance mounts, leaders must face the delicate balancing act of adopting a technology whose creators are increasingly unpopular.

How can businesses deploy AI while keeping a distance from the creepy AI industry?

I. Where are the gains?

The economic windfall from AI is late. Although most economies grew in 2025, hiring has generally slowed. Growth generated by investment rather than efficiency is unevenly distributed. In the US alone, 90% of growth is attributed to spending on data centres.

At the same time, adoption remains broad but shallow. While 41% of US workers used generative AI by late 2025, only 13% use it daily. Globally, whilst 14% of the world's population uses AI, a tiny 0.3% pays for it.

One explanation is that the noise is causing “organisational rewiring” to stall. Leaders are reluctant to redesign workflows and fully leverage AI’s capabilities until the dust settles.

II. A “botlash” is brewing

In the meantime, a “botlash” is brewing, as resentment converges across the political spectrum. Campaigns like “QuitGPT” and “Resist and Unsubscribe” now frame non-usage as a form of modern-day strike. Opposition is also moving to the physical world, as communities halted the construction of 20 data centres in just three months last year due to resource concerns.

Legislators are responding to this pressure. In the US, state legislators introduced over 1,200 AI-related bills in 2025 alone. The EU is opening multiple investigation fronts to curb the sector's excesses.

III. Diverging attitudes

Attitudes are diverging sharply by geography and age. Recent data highlights a fracturing consensus, with scepticism rising in developed economies and among younger cohorts. Nervousness is particularly acute in markets where white-collar stability feels threatened.

Simultaneously, whilst Gen Z most frequently uses generative AI, it is also the most anxious about its implications. This creates a workforce that is simultaneously dependent on the tools and resentful of the displacement they represent.

IV: A fine balancing act

For European business leaders, the path forward requires putting the technology to use, while insulating their brand from the backlash against its creators. This involves a number of subtle changes:

1. Know the supply chain. IT neutrality is untenable; leaders must keep an eye on the tech they use and the values it represents.

2. Augment, don’t replace. The backlash is fuelled by fear of job displacement, so adoption strategies must focus on augmenting teams.

3. Human-centric signalling. As tech giants alienate customers, there is SMEs can differentiate themselves by being aggressively human-centric.

Amodei’s “country of geniuses” paints a dystopian vision in which synthetic intellects rule and humans exist only on the margins. The public loves the tool but is beginning to hate the industry and its mesianic posturing.

Business leaders must walk this fine line, exploiting the opportunities of AI whilst keeping a distance from its creepy, tone-deaf creators.

Further reading

• Dario Amodei, The Adolescence of Technology

• The Economist, The AI productivity boom is not here (yet)

• Financial Times, The ‘botlash’ movement is gaining momentum

• Pew Research Center, How people around the world view AI

• Harvard Business Review, How Gen Z uses Gen AI – and why it worries them

“Nostalgia is not a strategy”, said Mark Carney in his widely lauded speech in Davos. As states force AI into walled gardens of geopolitical control, the IT neutrality of yesteryear has become equally untenable.

The evidence of diverging regulation is mounting. The US Defence Department is reportedly threatening to label Anthropic a “supply chain risk”. China continues to refine its domestic regulatory framework, prioritising national interests. Meanwhile, Europe is enforcing strict safety rules through the EU AI Act, and India champions a non-aligned “Global AI Commons“.

For European business leaders, the era of the borderless software stack is ending. Tech has a passport, and procurement is becoming an exercise in geopolitical risk management.

Read on to learn

1. The US, China, and the EU are developing incompatible AI rules

2. Companies should understand risk in their AI supply chain

3. Decentralised architectures preserve market access

4. Alternatives are emerging via partnerships and initiatives

5. CIOs should be ready to understand and operate as diplomats

I. The AI supply chain

Like other raw materials, business leaders must understand the provenance of their AI models. This requires a comprehensive audit of every model running within the business: who owns the intellectual property, where the data is hosted, and which legal jurisdiction governs its use.

Software buying is no longer just about performance but a declaration of alignment, designed to mitigate risk. Reliance on a US provider that falls foul of new EU rules could result in diminished access or fines. Deploying Chinese AI models may disqualify a firm from Western government contracts.

II. Decentralised architectures

The pursuit of a globalised AI infrastructure for a firm operating in multiple regions is increasingly hazardous. A more resilient approach involves a modular, region-specific architecture.

A firm might employ US-hosted models for North American operations to benefit from local speed and a different regulatory appetite. However, it may require distinct, localised models for Chinese operations to comply with strict domestic data and censorship laws.

This approach sacrifices some economies of scale, but preserves market access. The cost of maintaining separate systems is high, but the cost of being locked out of a major economy for non-compliance may be terminal.

III. European AI solutions

For operations within the EU, legal certainty is becoming a competitive advantage. With the EU AI Act’s full enforcement set to begin in 2026, the “Sovereign AI” alliance between SAP and Mistral AI illustrates a pragmatic solution.

In the current climate, sovereign AI is less an IT preference and more an insurance policy. By hosting European-engineered models on European infrastructure, firms can ensure data remains physically and legally within the bloc. This reduces exposure to the US CLOUD Act, which permits American law enforcement to access data held by US providers regardless of where it is stored.

IV. The Indian alternative

While superpowers fortify their digital borders, India is positioning itself as a neutral intermediary. At the recent summit in New Delhi, the government unveiled the “Global AI Commons,” a project intended to democratise access to computing power and datasets.

For non-sensitive, high-volume data processing, Indian partnerships offer a cost-effective route that avoids some of the legal entanglements of the US-China rivalry. India is positioning itself as a “back office” for the AI era: more affordable than the US and lower-risk than China.

V. Chief diplomacy officers

As the private sector becomes strategically important, the role of Chief Information Officers is evolving into that of a diplomat. Choosing an AI model is a strategic choice, as much as a technical one.

As digital borders harden, the most resilient companies will be those able to navigate multiple jurisdictions simultaneously. Sadly, IT neutrality is a relic of the past; deliberate, informed alignment is the way forward.

When ChatGPT emerged in late 2022, many observers were sceptical. They wondered if this was a genuine structural shift or another transient fad, like the metaverse or NFTs. In the years since, corporate AI adoption has often been performative. Pilots and MVPs have proliferated, frequently serving more as signals of innovation than as value drivers.

The focus now shifts toward business outcomes. Does the technology foster customer loyalty? Is there evidence of improved team efficiency? Does it contribute to the margin?

In other words: is AI creating value or just complexity? Consider using the following metrics as reliable indicators.

1. The 40% rule of product-market fit

A good indicator of product-market fit is that it’s considered a “must-have” by 40% of users. Especially useful for startups, AI allows firms to iterate rapidly before committing to expensive hires. This efficiency enables businesses to remain bootstrapped for longer, preserving capital and autonomy.

2. The LTV:CAC ratio

A healthy business should aim for a ratio of 3:1 or better between Lifetime Value and Customer Acquisition Cost. AI can be used to identify the specific go-to-market tactics that yield results with minimal spend. This ratio remains one of the clearest indicators of future profitability.

3. Revenue per employee

AI is designed to augment human effort rather than simply replace it. Equipped with the right assistants and agents, leaders should expect each employee to support roughly 20% more revenue year-on-year. Over the longer term, this should lead to a measurable reduction in the relative costs of management and administration.

4. Net Promoter Score

Customers are becoming impatient with poorly implemented AI. Leaders should resist the urge to cut headcount immediately. Instead, they should direct the time saved by automation toward higher-quality client interactions. The goal should be an NPS of 70% or higher, generally considered a mark of high customer satisfaction.

5. Employee churn

Fear of automation can undermine adoption and unsettle a workforce. Leaders must ensure that AI targets repetitive tasks without stifling creative or strategic work. If annual employee churn rises above 10%, it suggests the integration strategy requires a rethink.

6. The (other) rule of 40%

Companies at different maturity stages focus on different financial targets. Startups look for growth whereas more established companies prioritise profits. To make financial sense, leaders should ensure that the sum of annual growth and profit margin is 40% or higher, as AI adoption gathers pace.

Numbers After All

To be sure, most of these metrics concern humans: customers interacting with staff, and employees collaborating with one another. This human layer is often slow and complex. Rushed technological solutions can easily lead to unintended consequences.

For most companies a steady pace of adoption is the safest way forward. Leaders should keep an eye on all these metrics and ensure business fundamentals remain strong along the journey.

Further reading

1. AI-native companies are scaling 2-3x faster ICONIQ Capital

2. Measuring the “40% rule” of product-market fit PMF Survey

3. Gen AI is shifting the needle on customer satisfaction McKinsey & Co

4. The 2025 SaaS Metrics Benchmark RockingWeb Report

5. Using AI to maintain the “Rule of 40” MIT Sloan Management Review

It is hard to maintain a distinct company culture when everyone is working remotely. Growcreate, the digital solutions agency I co-founded 14 years ago, makes it look easy. Informal yet professional, technical yet customer-focused, we remain true to ourselves and to our clients.

As we gather in Prague this week for our annual summit, we are taking stock of a fast-moving landscape. Code and marketing are the areas most immediately impacted by AI. Clients expect us to use AI to expedite development, which puts pressure on billable hours. Learning a new tech stack will take time. In an overhyped, noisy marketplace, establishing credibility is hard.

Challenges are real for established companies, but so are opportunities.

1. Increased demand for support

No technology survives without being maintainable. Businesses evolve, and technology must be able to track their growth. To move from pilots to production, companies need access to mature and responsive support operations.

AI will automate some tasks, but at its core, technical support is intensely human-centric. Investigating edge cases, communication and negotiation are areas where humans will continue to excel. This plays to the strengths of companies like Growcreate, which prioritise support and iterative enhancements over one-off projects.

2. AI trains run on data rails

AI only delivers value when trained on large volumes of high-quality, interconnected datasets. Improving workflows with AI agents will only be possible if they can connect to multiple systems across departments and environments. This is challenging, both technically and organisationally.

In this environment, agencies with a broad technical orientation are well positioned to perform. Companies should look for partners who can integrate data, Cloud, and DevOps to ensure their AI solutions run on well-maintained data pipelines.

3. Exploration to exploitation

AI is finally boring; the flow of LLM releases is a mere backdrop to the real story: adoption. As focus shifts from exploration to exploitation, companies will need independent advisers and integrators they can trust. To design workflows that bridge departments, trust will need to be strong and durable.

To drive adoption, businesses will engage partners willing to build deeper and stronger relationships. Agencies that listen to customer needs and can work across departments will prosper.

We will continue to support, to enhance, to evolve

Growcreate was founded as digital transformation was reaching maturity. We disrupted web designers who did not see that our craft was moving from projects to ongoing support. That design was a subset of customer experience, and Cloud was not fancy hosting but agility.

Along the way, we built capability and the kind of staying power that most agencies only dream of. Our average client stays with us for 5 years, and 5 of them have been with us for over a decade. Our revenue is growing, recurring and diversified.

Businesses facing disruption need two things: the flexibility to adapt to the new world and the strength to navigate the transition. Few companies look forward to disruption — unless they are ready for it.

The evidence is mounting. “Can I finally get through to a real person?” asks the frustrated user. “Look at this—it just totally made it up!” exclaims another. Even early adopters report that AI features are becoming overexposed, hogging the limelight and obscuring utility.

Just as brands do not plaster “Hosted by Azure” or “Made with Tailwind” across their homepages, it is time for AI to recede into the background. While AI can power a responsive, personalised experience, customers do not want it to be the experience.

Resistance is building

Over the last few months, resistance to AI has been observed across multiple feedback channels. User testing, analytics and surveys reveal a negative reaction to explicit “AI features,” ranging from mild irritation to active hostility.

This stems from three perceptions:

• AI is a cost-cutting hack that blocks human access, over-automates customer service, and handles edge cases poorly. At some point, every customer will face an unusual situation, so the effect is felt across the board.

• AI produces poor results, confirming expectations of inaccuracies and hallucinations. Brands are held to a higher standard than free tools, and hard-won reputations unravel quickly.

• AI is not to be trusted, driving fears of harmful hallucinations and mass unemployment. With media coverage oscillating between the apocalyptic and messianic, mistrust is bleeding into user sentiment.

99 problems, AI is not one

Customers have multiple reasons to interact with digital platforms, none of which is AI. They want to access information, purchase a product or manage their account. AI can make these interactions more pleasant and personalised, but it should not overshadow the experience.

Consider the following friction points:

• Customers are tiring of typing. Chatbots and search boxes are taking up increasing amounts of screen real estate, yet users still prefer point-and-click and touchscreen interactions.

• Customers are becoming “slop aware.” Generated text and media are increasingly viewed as low-value fillers. While AI is an invaluable creative assistant, it should be part of the process, rarely the final product.

• Customers appreciate exit routes, such as a direct line to human agents or the option to hide AI content. Anecdotally, Google’s AI Overviews are widely adopted because they can be overridden via the Web tab.

AI-native means AI-inside

It is not a paradox that customers are enthusiastic about AI yet sceptical of how businesses employ it. Chat conversations are ephemeral; brand relationships are more permanent and demand higher quality.

Chat conversations are ephemeral; brand relationships are more permanent and demand higher quality.

Many businesses, keen to demonstrate an innovative edge, are becoming performative with AI. This risks creating a disconnect with customers who care about the utility and the experience, not about the technology behind it. Worse, AI overexposure can backfire, leading to a loss of trust and reputation.

AI will likely remain the main act in investor pitch decks for some time. But when it comes to products, it is time for AI to join the rest of the tech backstage.

In conversations with clients and partners, AI is often depicted as a comet hurtling towards Earth. Faced with a supermassive monolith, our only choice is to brace for impact. In this narrative, the negative effects are as unavoidable as the opportunities.

This is too fatalistic. While the eventual transformation of workflows and value chains is inevitable, business leaders have discretion over the manner and velocity of adoption. Companies can, and should, exercise control over how they deploy AI solutions.

Decision 1: Augment or cut jobs?

To follow the narrative that AI creates efficiencies by reducing headcount is to prioritise short-term gains over long-term competitive advantage.

Deploying AI to retain staff is a strategic choice. By automating mundane tasks and empowering your team to focus on high-value work, workers become more essential to the business, not less. Happier teams are more productive, leading to higher customer satisfaction and loyalty.

Decision 2: Momentum versus safety

The “move fast and break things” mantra of Silicon Valley is often ill-suited to mature companies, where reputations are hard-won and easily lost.

Balancing momentum with rigorous testing for safety, privacy and accessibility is prudent. Agility is important, but when the technology has the potential to mislead or leak personal data, not every iteration is fit for the public domain.

Decision 3: Environmental impact

Just because AI providers subsidise their services to claim market share does not mean environmental costs are gone.

There are practical ways to minimise the carbon footprint of AI solutions. This involves using LLMs only when necessary and opting for smaller, more efficient models where possible. Training teams to economise on energy-hungry prompts (particularly in video generation) and carefully selecting vendors are solid steps toward sustainable AI.

Decision 4: Limits to autonomy

Technological capabilities are always moderated for safety. Just as driving speed is limited well below the car’s capability, leaders must be careful to place limits on AI autonomy.

Allowing AI to handle specific, low-risk tasks is acceptable, provided a human remains in the loop to approve all non-reversible operations and final outputs. As regulation struggles to keep pace with technology, leaders bear a greater responsibility to enforce appropriate safety measures.

Decision 5: Vendor selection and US monopoly

The assumption that American tech giants are the only vendors of AI models and services should be challenged. For Europeans, total reliance on US companies risks losing jurisdiction and sovereignty.

Europe possesses a flourishing ecosystem of AI companies, particularly at the application layer. Supporting local vendors encourages entrepreneurship, innovation and helps develop the technology ecosystem. In an era of geopolitical uncertainty, vendor choice is a core component of risk management.

Mind your business

Struggling with basic AI adoption, some companies may find the burden of ethical decision-making too heavy. Others will disagree with these concerns or not deem them urgent enough.

The key takeaway is that AI is not a one-size-fits-all proposition, and that business leaders have the power of choice. Given its transformative power, how they exercise that power is a profound expression of their companies’ values and future direction.

Documentation, reports, specifications, emails… Modern businesses generate so much content that one might think it is the reason they exist. With few exceptions, producing this content is a chore that most humans will happily outsource to AI.

But as AI commoditises content, attention shifts to workflow design. AI agents can use applications to complete tasks with some degree of autonomy. But where do they fit in your workflow?

Dropping an agent into a broken process merely accelerates chaos. To deploy them effectively, we must first look at Workflow Design.

Key concepts

1. As content gets commoditised, focus shifts to workflow design

2. Use a framework like Dynamic Work Design for a structured approach

3. Visualise with n8n to expose gaps between ambition and reality

4. Find improvements in information disconnects between tasks

5. Use AI agents to fix and automate slow and repetitive processes

1. Dynamic Work Design framework

Developed at MIT, Dynamic Work Design (DWD) is a “process of ongoing, hands-on problem-solving” that helps teams replace chaos with a calmer, more effective workflow. It is grounded in common sense: once you understand its principles, you keep seeing applications everywhere.

1. Solve the right problem. Resist strategic reviews; look for small disconnects in high-impact areas.

2. Structure for discovery. Ensuring that measurements are in place to link targets to real-world metrics.

3. Connect the human chain. Examine how tasks are handed over, and precisely what information the next link needs.

4. Regulate for flow. Matching tasks to capacity to avoid backlogs and burnout.

5. Visualise the work. Use a system that illustrates the process, owners, capacity, and work in progress.

DWD is particularly helpful in customer-centric problems. Can we speed up onboarding from 5 days to 5 hours? Can 90% of service queries be responded to within 9 minutes? If you are unsure where to start, begin by visualising your current workflows.

2. Using a workflow design tool

The second ‘D’ in DWD is important: the framework is based on Design Thinking, which solves problems through iteration and visual representation. To help “visualise the work”, we can use an automation tool like n8n (Make and Zapier are also good).

Acting as a “Figma for workflows”, these tools allow us to map:

• Tasks of different types and scopes, connected sequentially or branching into conditionals and sub-workflows.

• Information requirements, or data properties that must be entered manually or automatically.

• Capacity tracking to monitor work items as they move through the system (what n8n calls “Executions”)

• Simulations that allow you to trigger tasks, monitor running time and test interventions.

To begin with, run n8n in “Design Mode”, using only manual tasks. Ensure your design reflects how tasks get passed from one person to another in the real world. Many companies reach the first “aha” moment here, as missing links and dependencies come to light.

3. Fixing information disconnects

The third rule of DWD (“Connecting the human chain”) relies on two specific states: the Handover and the Huddle.

Ideally, each task should contain enough information for the next person to begin immediately. This is a clean Handover. If the information is incomplete or ambiguous, the team will organise a Huddle. A Huddle resolves the issue, but it consumes time and capacity.

Take a look at your workflow and search for disconnects:

• When a task is handed over, are the recipient’s information needs clear to the sender?

• When a task requires a Huddle, is that meeting necessary, or is it just covering for a lack of documentation?

• Are there clear triggers to switch a Handover to a Huddle when exceptions occur?

• Is there sufficient capacity (redundancy) to handle spikes in volume?

Many disconnects will be resolved through communication. When team members understand how their work output is used upstream, they adapt. Others need management to rearrange tasks and capacity.

Rule-based automations can speed up many mundane tasks, such as data transformation, system integrations, and sending notifications. When all these interventions are in place, you should consider AI agents.

4. Automating with AI Agents

AI Agents are built on the intersection of automated workflows and LLMs. They excel where logic is required - territory beyond the reach of standard, rule-based automation. We rely on three key components: an LLM for reasoning, persistent memory for context, and tools for accessing and modifying data.

Using n8n, you can deploy agents to fix even the trickiest disconnects:

• Prepare content to facilitate cleaner Handovers and Huddles, such as release notes and product documentation.

• Simulate a recipient to verify that incoming data is sufficient, alerting the sender immediately, and saving the human recipient’s time.

• Act as a cross-domain “translator”, enabling a recipient to “chat” with incoming data, rather than struggling with technical content.

• Automate admin work, specifically data validation and entry, where speed matters and the outcome can be easily verified.

True to the probabilistic nature of LLMs, agents are flexible, but results will vary. For reliably accurate output, stick with rules-based automation. They require extensive data “plumbing” and testing before they can be safely deployed.

However, their ability to reason makes agents infinitely more powerful than rulesets. They adapt to specific situations where “one-size-fits-all” fails and can handle unpredictable datasets. Importantly, they can be programmed to learn and improve over time.

5. Closing: Get real with AI Agents

The AI industry, keen to keep the hype train on its tracks, has found a new hero in Agents. We are seeing instances of agent-washing, where simple automation is disguised as advanced AI. To unlock real value, ignore the technology and focus on your workflow:

• Adopt a Workflow Design framework: use Dynamic Work Design - or another framework - to visualise the workflow and uncover where errors and delays occur.

• Fix the disconnects: address problematic information flows by improving communication or implementing rule-based automation.

• Deploy AI agents to accelerate: create further efficiencies with agents that can reason and act: “translating” docs, simulating interactions and handling admin.

Five years ago, describing AI content generation as “game-changing” would have earned you a few confused looks. Yet, the sheer volume of documentation created by companies has already transformed the workplace.

We are now ready to tackle teamwork and coordination. Accelerating workflows will impact value chains, supply chains, and entire economies. As with most generalist technologies, we will overestimate their short-term impact and underestimate their long-term effects.

Beyond the nasty (Grok) and the hearsay (DeepSeek), AI is quietly embedding itself in everyday life. SMEs can find opportunities in healthcare, commerce, and email.

1. OpenAI launches Health app

Need to know: Responding to its own findings of 40 million health-related prompts per day, OpenAI has introduced ChatGPT Health, a specialised service designed to handle sensitive medical data. It integrates with apps such as Apple Health and MyFitnessPal, but the data appears to be isolated and never used to train OpenAI’s models.

Need to act: AI gets embedded in the healthcare ecosystem. If you operate in this space, consider the implications for patient safety, diagnosis, care delivery, and health insurance. It is a highly sensitive space, ripe for innovative approaches and ideas.

2. Copilot offers a POS terminal

Need to know: Microsoft has launched Copilot Checkout, a feature that lets users discover and purchase products directly in the chat window without being redirected to a retailer’s website. Partnering with PayPal and Shopify, the system aims to reduce friction.

Need to act: AI gets embedded in the commerce ecosystem. As we move closer to “agentic commerce”, brands need to reconsider how to retain share of attention and avoid commoditisation.

3. Gemini vs. Copilot fight for your email

Need to know: The battle for your inbox ignites as Google upgrades Gmail, packed with Gemini 3 goodness: “AI Overviews” to summarise threads and answer inbox questions, “AI Inbox” to filter noise and prioritise critical updates and a Grammarly-style proofreader.

Need to act: Building on the success of AI Search, Google is integrating its AI productivity suite. At the time of writing, it is probably the strongest product in the market and could challenge Office even in larger organisations.

Adam Smith published The Wealth of Nations, widely regarded as the most important book in economics, in 1776. Though it is claimed that its significance exceeds its quality, its key concepts (the Invisible Hand, Division of Labour and Self-Interest) remain timeless and make intuitive sense a quarter of a millennium later.

Our world would be utterly unrecognisable to Adam Smith. His thinking was foundational to the Industrial Revolution, but production lines did not materialise until a century later. The sheer speed of modern markets would be beyond his wildest imagination. But like all classics, the book’s core concepts remain relevant and offer useful insights for thoughtful leaders.

Key concepts

1. AI reframes Adam Smith’s 3 key concepts of the free market, with profound implications

2. The Invisible Hand operates in real time, shortening planning cycles

3. AI agents will require a rethink of the Division of Labour and workflows

4. Sharper Self-Interest will encourage micro-enterprises and outsourcing

5. SME leaders need to plan for hyperactive and busy marketplaces

1. The bionic Invisible Hand

Business strategy is so closely connected to the concept of the Invisible Hand that it’s generally taken for granted. A change in pricing or an innovative new product creates new revenues, because buyers in a free market will purchase it, as if “led by an invisible hand”.

The Invisible Hand is about market agility. As competitors change their stance, leaders must continue to adapt to make sure the Hand moves in their favour. AI is set to accelerate business intelligence and execution strategies:

• Dashboards will be replaced with conversations, helping leaders gain insights faster and more intuitively.

• Decision-making will be faster with AI providing instant analysis, predictions and simulations.

• AI agents will collaborate on open standards, instantly updating partners and customers of upcoming changes.

As faster decision-making turns the Invisible Hand bionic, planning cycles will shorten from years to months. In response, leaders should prioritise agility in product design, marketing and distribution.

2. Division of Labour gets upended

Adam Smith famously cited a pin factory to illustrate the concept of the Division of Labour. A dozen specialists, each responsible for a single task in the workflow, can scale production to a level that no single individual could match.

This concept, so fundamental to education, employment and management, is about to be upended. With AI agents providing cross-domain assistance, team roles will become functionally broader and more specialised in specific contexts.

• Hybrid roles will emerge, relying on AI Agents to operate confidently across product, marketing, sales, and legal functions.

• Specialisations will be vertical, focusing on specific contexts: industries, segments, channels or key accounts.

• Workflows will be horizontal, teams self-organising to deliver customer-driven outcomes.

By automating low-level tasks, AI not only saves time and effort but also redefines how we divide work. As you consider deploying AI Agents, prepare to redesign your workflows to make the most of their capabilities.

2. Self-interest powers ecosystems

It is not out of “benevolence” that employees dedicate their energy and employers provide jobs, but out of their respective self-interest. The balance works, even when self-interest includes non-monetary factors such as the work environment and work-life balance.

However, in faster-moving markets and teams with dynamic and horizontal workflows, self-interests will be harder to reconcile. With a low barrier to entry, AI will enable many professionals to offer their expertise from independent vehicles.

• Seeking efficiencies from AI adoption, many employers are expected to reduce full-time headcount.

• Experienced professionals will use AI to set up as freelancers and micro-businesses.

• To grow, companies will rely more on outsourcing and ecosystems of trusted partners and vendors.

Like the SaaS wave before it, AI will enable a new wave of entrepreneurship and micro-enterprises. Ambitious SMEs will need to outsource many business functions and should be ready to review their Make-or-Buy strategies.

4. Summary: a hyperactive market

The post-AI marketplace will be fragmented and hyperactive, characterised by numerous participants, frequent directional shifts, and noisy competition. Some organisations will thrive, while others will rely on legacy competencies.

Whilst some of these developments are already a reality, others will take years to materialise and are facing obstacles such as technology adoption, social attitudes, and market corrections. Nonetheless, forward-thinking SME leaders should take these into account when planning for the long run.

Recommendations for SMEs

• The Invisible Hand goes bionic
  Faster decision-making will shorten planning cycles from years to months. Leaders should prioritise agility in design, marketing and distribution.

• Division of Labour gets upended
  AI agents will broaden team roles, and teams will specialise horizontally. SMEs should redesign their workflows to better meet customer needs.

• Self-interest powers ecosystems
  As AI encourages entrepreneurship, SMEs will increasingly rely on outsourcing. You should revisit your make-or-buy and partnerships strategy.

5. Closing thoughts

We will never know what Adam Smith would make of our world, but it is reassuring to know that ideas conceived 250 years ago remain relevant. It reminds us that, despite technological advances, concepts such as free markets, specialisation, and incentives remain timeless. Like a good book, we can always turn to them to make sense of a world that seems to spin ever faster.

My 2026 predictions sketched out the overflow of the AI hype cauldron and its aftermath. But how did we get here? What created these circumstances, and how are they connected?

In my final post of 2025, I present the 5 key numbers that summarise its biggest stories. Whilst not really strictly scientific, they illustrate our mixed feelings for this year: exuberance, uncertainty, and a sense of impatience about what lies ahead.

5 trillion dollars

The T-Rex of the big scary AI numbers, $5TN, is both the peak valuation of Nvidia, and the CapEx that Big Tech has earmarked for AI data centres until 2030, according to McKinsey. Depending on who you ask, there are 5 trillion reasons we are in a remarkable boom or a precarious bubble.

30X cheaper

The release of DeepSeek R1 in January made a big splash and even a bigger dent in Ndivia’s stock, thanks to a new approach and clever engineering. To understand why the frequently cited 30X cheaper is a game-changer, consider that it represents a 97% cost reduction relative to the competition.

95% pilots failed

Perhaps the most-cited figures of the year: 95% of all AI pilots have failed to deliver ROI, according to MIT. Though the methodology is dodgy, the noise it generated spoke volumes. Commentators jumped on it because it supported our collective observation that business adoption remains underwhelming.

55K jobs lost to AI

Although evidence of AI’s impact on jobs across economies is limited, the announcements are sufficient for CNBC to estimate that AI eliminated 55,000 jobs, mostly in tech. With vibe-coding valuations reaching for the stars, it seems that software developers are on their way to making themselves redundant.

50% of web content is AI

Slop, rage-bait, parasocial - the words of the year indicate our impatience that AI is making everything, everywhere, all at once. Whilst the figure that half of all internet content is AI-generated is speculative, there is no question that we are overindulging. Keep it up, and the word of 2027 could well be “model collapse”.