Everyone assumed AI adoption would go one way. It went the other way.

The consensus story for the last two years has been that AI is spreading through enterprise faster than any technology in history. The reality is almost the opposite. According to the US Census Bureau's Business Trends and Outlook Survey, only 17-20% of US businesses actually use AI in a business function. Four out of every five American companies are still, in effect, looking around for their shared Google Drive logins.

And it gets worse. MIT's 2025 NANDA study put a number on how bad the disconnect is. Despite $30-40 billion in enterprise AI investment globally, 95% of corporate AI pilots deliver no measurable impact on P&L. Only 5% actually reach production.

The predictions everyone made about why this would fail turned out to be wrong.

It's not the model quality.

It's not the regulation.

It's not that enterprises are too slow.

The specific bottleneck is that almost nobody inside these organisations knows how to actually wire the tools into the work.

For the AI industry itself, this is a genuine problem. For tech-savvy founders watching from the outside, it's the biggest short-term opportunity of the cycle. Because the gap between what enterprises need and what the labs can deliver on their own is now too wide for Anthropic, OpenAI, or any single vendor to close by themselves. That gap is exactly where the money is flowing right now.

See below to continue…

Your network is scattered across inboxes, calendars, messages and apps. SoPhy pulls it all together - and lets you access it the way you already communicate.

Ask it anything about anyone, straight from iMessage, WhatsApp or Telegram. It knows who you met, when you last spoke, and when to follow up.

The first integrated version launches late August. Waitlist members get first access.

Join the waitlist

Continue…

The MIT data names the bottleneck precisely. Companies that partner with external implementers see a 67% success rate. Companies that try to build AI implementations internally see a 33% success rate. Twice the failure rate, half the outcome.

The reason is unglamorous. Enterprises have decades of legacy processes, siloed data, undocumented workflows, and internal politics around who owns what. Frontier AI models are extraordinarily capable in the abstract and extraordinarily fiddly to embed inside those specific realities. Someone has to sit inside the organisation, understand the actual process, design the specific integration, build the connectors, handle the change management, and iterate for months until it works.

That work is nobody's core competency yet. The AI labs are not built for it - they build models, not implementation projects. Traditional consulting firms are only just starting to acquire the skills. Enterprises themselves cannot recruit fast enough - according to an executive search firm cited on the Frontier podcast recently, there are approximately 2,000 engineers in the US who genuinely know how to deploy AI inside a large organisation. Not 2,000 available. 2,000 in total.

That is the bottleneck. That is the gap. And it is currently far too wide for the AI labs to close by hiring alone.

The frontier labs know they cannot solve this themselves. Their behaviour over the last three months makes that unmistakable.

On July 15, Anthropic launched Ode - a $1.5 billion joint venture backed by Blackstone, Hellman & Friedman, Goldman Sachs, and Sequoia Capital. Not a product company. A services company. Its entire business model is sending small teams of senior engineers into large enterprises to wire Claude into their workflows. Five weeks after launching, Ode announced its second acquisition - Casper Studios, a four-year-old consultancy with about a dozen technical staff whose client list included Netflix, Pepsi, hedge funds, and private equity firms.

OpenAI is doing the same thing. Its enterprise services vehicle, The Deployment Co., has raised $4 billion at a $10 billion valuation from TPG and Brookfield. Thrive Holdings, backed by Thrive Capital, raised $2 billion for the same purpose.

Read that back.

If the US market has 80% of companies not yet using AI meaningfully, the European market is almost certainly worse. UK enterprise AI adoption typically runs behind US figures by six to twelve months. The gap between demand and supply is wider here than there.

Meanwhile the European labs, systems integrators, and traditional consultancies are moving even more slowly to close that gap than their US counterparts. The specific window that Anthropic and OpenAI are racing to close in the US is much wider open in Europe, and much less contested.

There are three genuine routes through this window right now.

The least glamorous route and the most immediately profitable. There are founders in our community who have pivoted their businesses toward AI implementation consulting in the last twelve months. One of them made £1 million in six months, sitting inside corporates and doing exactly what the MIT study describes as the missing piece.

This is not what most ambitious founders imagine themselves doing. It is also, right now, the fastest cash flow available to anyone with genuine AI capability and enterprise credibility. And it has a strategic benefit most founders miss: consulting inside 15 to 20 corporates over 18 months gives you exactly the enterprise relationships and specific problem discovery that a product business would take five years and £3 million in sales spend to develop. Palantir has been running this playbook for two decades under the name "forward-deployed engineering."

The 95% failure rate is not only a services opportunity. It is a product opportunity.

Every enterprise struggling to deploy AI needs the same set of things. Connectors that plug frontier models into legacy systems without breaking them. Interfaces that non-technical staff can actually use. Evaluation and monitoring tools that show whether the AI is delivering value. Data preparation layers that turn messy internal knowledge into something models can retrieve. Governance and safety infrastructure that makes IT departments comfortable signing off.

Almost none of this exists in the shape enterprises actually need it in. Founders who can build products that solve these problems - and that a non-technical enterprise buyer can actually understand and buy - are entering a market with measurable, budgeted demand. The advantage over pure services: it scales. The disadvantage: it takes longer to first revenue.

The specific version of the adoption problem everyone will be dealing with in 12 months is AI agents. Frontier labs are pushing agentic AI as their next frontier, which means enterprises are about to start trying to deploy autonomous AI systems that take actions on their behalf. The trust, security, containment, and audit infrastructure required to do this safely is essentially not built yet.

The same enterprise buyers who cannot get their basic AI pilots to work in production are about to be asked to deploy autonomous agents in production. The infrastructure gap for agents is going to be an order of magnitude larger than the current implementation gap. Founders building in this layer now will be genuinely early.

Two things are actively narrowing this opportunity.

The frontier labs are systematically buying up the supply. Every consultancy Anthropic, OpenAI, and their equivalents acquire is one less independent implementation shop in the market. When Ode acquired Casper, it took a proven twelve-person team off the table for every competing vendor. That consolidation will accelerate.

Meanwhile, the specific patterns that make AI work in production are being identified, documented, and diffused. The MIT study itself is part of that diffusion. In two years "AI implementation" will be a standard capability inside the big four consulting firms and the systems integrators. Right now it is a genuine specialist skill.

The specific opportunity to be one of the small number of people or companies in Europe that can bridge this gap is not open forever. It is open for approximately the next 18 to 24 months.

This is not the invention that colonises Mars. This is not the next Anthropic. This is not a defensible platform business over ten years.

But it is, right now, one of the largest and most specific gaps between what a market urgently needs and what its current vendors can deliver. It has cash flow implications most product businesses don't have. It has enterprise relationship implications most founders would work for years to develop. And it has a genuine expiry date.

For founders in Europe with technical AI capability, existing enterprise credibility, or the ability to build products that make AI meaningfully easier for non-technical buyers to adopt - this window is worth thinking about seriously in the next quarter. Not forever. Not as your identity. Just as a specific strategic move that can fund longer-term ambitions and build the exact enterprise access your future company will need.

The AI story everyone got wrong is that adoption would happen quickly and evenly. It isn't. The specific unevenness is where the money is.

The founders who see it will not remember this period as the moment they got distracted by consulting. They will remember it as the moment they built the enterprise credibility, the cash reserves, and the specific product insight that funded everything that came after.

The ones who don't see it will spend the next 18 months watching more pragmatic peers close revenue and build relationships they could have built themselves.

✅ Know a founder with strong AI skills who's grinding on product with no revenue while enterprises around them are desperate for exactly what they can do? Forward this their way. The window on this specific opportunity is real, and it doesn't stay open forever.

(👉 Vote now - we’ll share the results in next week’s issue. All votes are anonymous.)

(Last week’s poll results)

Access here and please share it with others.