In short
- The hardest part of implementing AI is no longer the model. It is finding people who can make AI work inside a real business process.
- That is why Anthropic is putting $100 million into training 10,000 engineers. European businesses that considered AI but did not adopt it name lack of expertise as their main reason.
- Decide deliberately how you organise that expertise: train your own people, have it built, or build together with a handover. Whichever you choose, you need an internal owner.
Organisations that want to implement AI rarely get stuck on the technology any more. Good models are available to everyone. The bottleneck is expertise: people who understand how your business runs, what AI can do reliably, and how to connect the two in a working process. That combination is scarce, even at the world's largest companies.
That does not mean you have to hire an AI specialist first. It does mean you have to choose how you organise that expertise, and someone inside your organisation has to own it. Below: what is happening, what knowledge you need, and which route fits your situation.
What did Anthropic announce?
On 2 October, Anthropic, the company behind Claude, launched the Claude Frontier Academy. It is committing $100 million to train 10,000 Frontier Deployed Engineers by the end of 2027: engineers who can take AI from an idea to a system in production. The announcement opens: "Today we are launching Claude Frontier Academy to solve one of the most pressing issues in AI implementation: talent" (Anthropic).
The programme starts with a four-day intensive in which participants build a Claude system for a simulated enterprise. Those who pass then lead a real deployment in their own organisation for twelve weeks, supported by Anthropic. The first cohorts come from firms such as Accenture, Deloitte, McKinsey and Novo Nordisk, and run in San Francisco, New York and London.
For most mid-sized organisations, then, this programme is out of reach. The signal still matters. The maker of one of the strongest models is effectively saying: our model is not the problem; the shortage is people who can put it to good use.
Why is the bottleneck people?
Anthropic bases this on what it sees at its customers: "a small group of deeply skilled people drives an outsized share of what AI delivers." That is the observation of a vendor with a commercial interest in more deployments, but independent figures point the same way.
In 2025, 20 percent of EU enterprises with ten or more employees used AI: 17 percent of small firms, 30 percent of medium-sized firms and 55 percent of large ones. Among enterprises that considered AI but did not use it, the most common reason was lack of relevant expertise (71 percent), ahead of unclear legal consequences (53 percent) and data protection concerns (49 percent) (Eurostat). Dutch figures tell the same story: 73 percent of Dutch firms in that group cite lack of experience (CBS).
The gap between small and large firms stands out. One likely explanation: larger companies more often have people who can free up time to work out how it all fits together. Smaller organisations usually do not.
What expertise does an AI implementation need?
An implementation that works combines three kinds of knowledge. They are rarely found in one person.
- Process knowledge. Who does what, where the exceptions are, what counts as good enough. This lives inside your organisation and cannot be bought in. Anthropic explicitly names "a deep understanding of how their business runs" as a condition.
- AI fluency. Knowing what a model does reliably, where it goes wrong and how to check its output. This can be learned.
- Technical delivery. Integrations with your systems, security, permissions, maintenance after launch. This is specialist work, and for most organisations not a full-time role.
If one is missing, things stall. A prototype built by an enthusiastic employee without a connection to your systems stays a demo. A technically strong solution that nobody from the process helped shape does not get used.
Which route suits your organisation?
There are broadly three ways to organise that expertise.
| Route | Fits when | Watch out |
|---|---|---|
| Train your own people | You want AI used widely in daily work and people have time for it | Training alone rarely delivers integrations and maintenance |
| Have it built | One defined process needs to work well, quickly | Without a handover you stay dependent on the builder |
| Build together, then hand over | You want results and in-house knowledge at once | Needs time from an internal owner during the project |
Every route needs an internal owner: someone who knows the process, decides what is good enough and keeps it up to date after launch. Without an owner, AI remains a side experiment, however good the training or the builder.
Where do you start?
- Pick three to five processes with a lot of manual work, waiting time or errors. Start from the bottleneck, not from a tool.
- Name an owner for each process, with time set aside and a clear reporting line.
- Agree the ground rules: which tools and licences, what data may and may not go in, who checks the output.
- Run one process for ninety days, with a measurable goal and a fixed review point.
- Only then choose the route: what will you learn yourselves, what will you have built, what gets handed over?
This is also how our leadership session on introducing AI in your organisation is structured. It produces a prioritised list of opportunities and a plan for the first ninety days.
When do you not need an AI specialist?
If your team mainly uses AI to write, summarise or research, solid basic training is often enough. Nothing needs to be built or integrated. In that case, start with Getting started with AI and agree what may and may not go into a tool.
And do not rush to hire before you know which processes you want to improve. A specialist without a clear brief will go looking for use cases, which is exactly the wrong order. Costs and returns vary widely by process; there is no general figure to give.
What is a good next step?
Map the three kinds of expertise for one process. Who knows the process, who can use AI well, who handles the technology? Whatever is missing determines your route.
At Hello Growth, strategy, build and activation sit in one team, and we build alongside the people who know the process. Want to work out which route fits your organisation? Book a conversation.
Sources
- Anthropic, Claude Frontier Academy: $100M to train 10,000 engineers, 2 October 2026
- Eurostat, Use of artificial intelligence in enterprises, data extracted December 2025
- CBS (Statistics Netherlands), Bedrijven gebruiken AI vaakst voor marketing of verkoop, 12 December 2025 (Dutch)
