Why 81% of Companies Are Stuck on AI — and the Three Things the 19% Do Differently

Two stories landed 24 hours apart this week. They're the same story, told from opposite ends of the table.

On May 4, Anthropic announced a $1.5 billion joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman to embed engineers directly inside enterprise customers. Hours later, OpenAI announced "The Deployment Company," a $10 billion venture with TPG, Bain, Brookfield, and Advent — and quietly attached a clause guaranteeing the private equity backers a 17.5% annual return for five years.

The next morning, Microsoft published its 2026 Work Trend Index. The headline number: only 19% of companies are operationally ready for AI.

Read together, those announcements are not three separate news items. They are one admission: the model is no longer the bottleneck. The organisation around the model is.

What The Labs Just Quietly Admitted

For two years, the bet at the frontier labs was that better models would carry the day. Make Claude smarter. Make GPT cheaper. Watch enterprise revenue follow.

That worked, partially. OpenAI's CFO Sarah Friar told CNBC in February that enterprise has grown to over 40% of OpenAI's revenue and is on track to reach parity with consumer by year-end. Anthropic has made similar moves — Frontier Alliances with BCG, McKinsey, Accenture, and Capgemini, plus its own Applied AI team embedded with customers.

But the May 4 announcements changed the shape of the bet. The labs are no longer just selling tokens to enterprises and partnering with consultants on the side. They are building (and in some cases, buying) consulting firms of their own. Reuters reported the day after the announcement that OpenAI's new venture was already in advanced talks on three acquisitions of AI services firms.

The why is easy to read in the financial structure. Fortune captured it in one line: "For every dollar companies spend on software, they spend six on services." If you build the model and stop there, you collect the dollar and watch six more flow to whoever shows up to make the model actually work inside the customer's business. The frontier labs decided they would rather be that other party.

The Palantir-style forward deployed engineer is the unit of delivery. Both ventures plan to put engineers inside customer offices, redesigning the workflows the AI runs on. Jon Gray of Blackstone called engineer scarcity "one of the most significant bottlenecks to enterprise AI adoption." That's the bottleneck the $11.5 billion is being deployed against.

The Microsoft Data Explains Why

If you wanted proof that the bottleneck has shifted, Microsoft handed it to you on May 5.

The 2026 Work Trend Index surveyed 20,000 knowledge workers across 10 countries between February and April. It scores companies on two axes: individual capability (how well workers actually use AI) and organisational readiness (whether the company is built around how AI changes work). The result is a five-tier classification:

Frontier — 19%. High individual capability and high organisational readiness. Both sides reinforce each other.
Blocked Agency — 10%. Skilled people, lagging organisation. Talent constrained by the structure around it.
Unclaimed Capacity — 5%. Organisation is ready, workforce hasn't caught up.
Emergent Zone — 50%. Both individual and organisational dimensions still developing. The middle.
Stalled — 16%. Low capability, limited organisational support.

Look at where the mass sits. Half of all companies are in the Emergent Zone. Another 16% are stalled outright. Only one in five has built the operating model that lets AI actually produce results.

The single most quoted finding from the report tells you why the gap exists. Microsoft's data shows that organisational factors drive 2x more AI impact than individual factors — 67% vs 32%. Culture, manager support, talent practices, and workflow design outweigh the worker's mindset by more than two to one.

Or, in Jared Spataro's framing for Microsoft: "Access to AI won't be the advantage for much longer. How the work is designed around it will be."

That sentence is the bridge between the two stories. The labs are racing to be the ones designing the work around the AI inside their customers' organisations, because they can see — in their own deployment data — that the customers can't do it on their own.

The Frontier 19% Doesn't Just Use AI Better. It Works Differently.

The interesting thing in the Microsoft report isn't the headline number. It's what the Frontier tier actually does.

80% of Frontier Professionals say they produce work they couldn't have created previously, compared to 58% for general AI users. They're not just faster at the same outputs — they're producing different outputs.

They also work differently. The report identifies four collaboration modes Frontier Firms use deliberately:

Author — humans produce, AI assists selectively.
Editor — humans set intent, AI drafts, humans review.
Director — humans spec the work, AI executes independently.
Orchestrator — multiple agents run in parallel, humans handle exceptions.

And the behaviour that separates them most clearly: 53% of Frontier Professionals pause to deliberately allocate which work goes to humans and which goes to AI, versus 33% of everyone else. 43% deliberately avoid AI on certain tasks to keep their own skills sharp. 26% document their workflows so the rest of the team can replicate them.

This is operating-model work. It's not "use the tools more." It's "redesign how decisions get made around the tools."

Why Only 26% Of Leaders Are Aligned On AI Strategy

One stat in the Microsoft report deserves its own moment. Only 26% of AI users say their leadership is clearly aligned on AI strategy. Only 13% of firms reward AI-driven workplace reinvention.

That's the friction point for most SMBs. Workers can see the upside. Leaders can see the upside. But neither has been given a clear direction on what to redesign first, what to leave alone, and what good actually looks like — so most teams default to bolting AI onto existing workflows and hoping the gains show up.

The Microsoft report calls this the Transformation Paradox: 65% of workers fear falling behind without AI, but 45% still find it safer to focus on current goals than to redesign work. Fear of standing still and fear of changing the wrong thing, in the same person, at the same time.

That paralysis is exactly what the frontier labs are now charging consulting fees to break.

The Three Things the 19% Do Differently

If you run a small or mid-sized business, the temptation reading all of this is to assume the action is at the top — that the OpenAI/Anthropic story is for Fortune 500 buyers and the Microsoft data is for HR leadership at scale. It isn't.

The Microsoft tiers map onto SMBs cleanly. Most are in the Emergent Zone. Some are stalled. A small number are pulling ahead — and the gap is widening fast.

Three things actually move you up the tier ladder, and none of them require an $11.5 billion budget:

Pick the work, not the tool. Most SMBs adopt AI tool by tool. Frontier Firms adopt workflow by workflow. Identify three high-frequency, decision-heavy workflows in your business. Those are the redesign candidates. The tool selection comes after.
Name the orchestrator role. Microsoft's data shows the differentiator isn't who uses AI — it's who decides where AI fits and where it doesn't. That's an orchestration role, and most SMBs haven't named it. Until someone owns the decision of what work goes to humans and what goes to AI, the work just gets done both ways and nobody learns anything.
Document one workflow this month. The behaviour that most cleanly separates Frontier Professionals from everyone else is documenting how they actually got the result. Pick one workflow you've redesigned with AI. Write down what you changed, what you kept, what broke, what you learned. That document is the start of an operating-model upgrade for the rest of the business.

None of this is glamorous. It's process work, decision-rights work, and writing things down. It's also exactly what the 19% are doing — and exactly what the frontier labs just bet $11.5 billion that the other 81% will pay them to come in and do.

The Honest Read

You can read the May 4 announcements as the labs maturing into full-stack AI providers. You can also read them as an admission against interest. The companies with the best models in the world are now telling the market that the models alone won't get the job done.

That's a real signal. Tooling is no longer the bottleneck. Operating-model design is. The companies that figure that out for themselves — and a small but growing number already have — won't need to hire a forward deployed engineer in 2027.

The rest will. And there is now $11.5 billion of capital, plus a 17.5% return clock ticking on at least part of it, making sure that engineer arrives on time.