The bottleneck isn't the model. OpenAI just spent $150m proving it.

The biggest gain available from AI right now is not a better model. It is whether the people in your firm can use the one you already have. OpenAI made that argument with its chequebook on 14 June, committing $150m to a new partner network and a plan to certify 300,000 consultants, on the stated grounds that model capability is no longer what holds organisations back. For a firm that is not a multinational, the roster of partners matters less than the diagnosis behind it, because the diagnosis should change where your next AI pound goes.

What OpenAI actually bought

Strip out the channel-programme detail and one sentence carries the news. "The limiting factor for seeing value from AI in the enterprise is no longer model capabilities," OpenAI wrote in announcing the network. "Instead, it's how organizations repeatably identify the right use cases, redesign workflows, integrate with existing systems, and drive adoption and change management at scale." The launch partners are the large consultancies, Accenture, Bain, BCG, McKinsey and PwC among them, and the money is aimed at training people to do that adoption work rather than at the model itself. When the company with the strongest commercial reason to talk up its models instead spends nine figures on the layer above them, that is worth reading as a signal.

It is also a signal a smaller firm can act on without buying anything OpenAI's partners sell. The diagnosis applies whether you are a bank or a 30-person advisory practice. The response does not have to.

The procurement reflex

Most firms meet an AI plateau with what we will call the procurement reflex. Results from the first tools flatten, so the instinct is to buy the next thing: more licences, a newer model, an agent platform. It feels like progress because it is a decision a leader can make in an afternoon and put on an invoice. It rarely moves the result, because the thing that stalled was almost never the tool.

The numbers behind the plateau are not subtle. In Logicalis's 2026 survey of CIOs, 89% described their organisation's approach to AI as "learning as we go" and 62% admitted to compromising on governance because they lacked the knowledge to do otherwise. Appetite was not the missing piece; 94% wanted to invest more. The gap sits between buying AI and knowing how to run it. A second tool does not close that gap. It widens it, because now there are two things no one has learned to use properly.

The skill that is actually scarce

What closes the gap is unglamorous and human. Ethan Mollick, who studies how people actually work with these tools, argues that the abilities that matter most with AI are management skills, the ones often dismissed as soft: scoping a problem clearly, saying what good looks like, giving useful feedback, and recognising when an output is quietly wrong. People who already do that with junior staff tend to get usable work from a model. People who cannot get confident nonsense, faster.

This is easy to see in judgement-heavy work. Take a first-pass diligence read on a target company. A corporate finance team that hands the data room to a model and asks for "the risks" gets a tidy, plausible list that an analyst then has to unpick line by line, often slower than starting cold. The same team, led by someone who can specify what a real red flag looks like in this sector, which contracts to weight, which revenue patterns to distrust, gets a first draft worth editing. Same model. The difference is the instruction, and the instruction is a function of the person, not the software.

The same pattern shows up in an accountancy team drafting client reports, or a recruiter screening a longlist. The firms getting value are not the ones with privileged access to a better model. Everyone has roughly the same models now. They are the ones whose people have learned to direct them.

What a smaller firm does instead

None of this requires the engagement OpenAI's launch partners are built to sell. A small advisory firm does not need an enterprise change programme. It needs a few people who are genuinely good with these tools and a leader who treats that as a capability to build rather than a box to tick.

In practice that means three things, and they are genuinely different from each other. First, pick the two or three real tasks where AI could change the work, not the demo tasks that show well in a meeting. Second, have your stronger people develop a way of doing each one that a colleague can copy, so the skill lives in the firm and not in one person's head. Third, put a single person in charge of how the firm uses AI, the way you would for any other shared standard, so decisions about tools, data and limits are made on purpose. That costs less than another year of licences, and it is the part the consultancies charge the most for, because it is the part that works.

This is the ground the Good Transformer Lessons for Leaders sessions are built on: starting from the work a leader and their team already do, and turning it into a few repeatable ways of working that keep judgement in human hands. The aim is not AI literacy in the abstract. It is your people getting reliably better results on the tasks that matter to the firm.

So before the next renewal or the next pilot, ask the harder question. If a sharper model arrived tomorrow, would your firm get noticeably more from it than it does today? If the honest answer is no, the constraint was never the model, and a new one will not move it. The work is to build the judgement to use what you already have, and it starts with the people who lead it. If that is worth an hour, book a discovery call.

FAQ

Why isn't our AI investment delivering results?

Usually because the constraint is adoption, not the tool. Firms that plateau have typically bought capable models but not built the workflows, judgement and habits to use them on real work. OpenAI's own position, set out when it launched its partner network in June 2026, is that model capability is no longer the limiting factor for enterprise value; the work of redesigning how tasks get done is. Buying another tool rarely fixes a problem the first tool did not cause.

Do we need to hire a big consultancy to get value from AI?

No. Large consultancies are useful for enterprise-scale change, and they are who OpenAI's partner network is built around, but a smaller firm can build the same capability directly and far more cheaply. Pick the few tasks where AI could genuinely change the work, have your strongest people develop a repeatable way of doing each, and put one person in charge of how the firm uses AI. That is the part that produces results, and it does not need a seven-figure engagement.

What skills actually matter for getting good work from AI?

Management skills more than technical ones. Ethan Mollick's work points to the abilities often called soft, scoping a problem, saying clearly what good looks like, giving feedback and spotting when an answer is wrong, as the ones that separate people who get useful output from people who get fast, confident errors. If someone is good at directing a capable junior, they tend to be good at directing a model.