Claude Frontier Academy: What Stays When the Engineers Leave
Ricardo Argüello, October 19, 2026
CEO & Founder
General summary
On October 2 Anthropic announced Claude Frontier Academy: $100 million to train 10,000 Frontier Deployed Engineers by the end of 2027, with engineers from Accenture, Bain, Capgemini, Deloitte, McKinsey and Morgan Stanley in the first cohorts. Training one takes months and runs by nomination, so most companies will rent this capability first. Whether the rental was worth it depends on what is still in your hands when the engineer leaves.
- Claude Frontier Academy was announced on October 2, 2026: $100 million, 10,000 engineers by the end of 2027, cohorts in San Francisco, New York and London.
- The program is a few days in person plus a 12-week residency inside the engineer's own company, where each participant leads a named Claude project. The first credentials are expected in early 2027.
- The first cohorts mix consultancies (Accenture, Bain, Deloitte, McKinsey) with companies training their own people (Commonwealth Bank, Morgan Stanley, Novo Nordisk).
- Mark Ajzenstadt put the renting risk in one line: what matters is what stays when the engineers leave, so rent from someone who plans to hand it over.
- Five things should end up in your company: the accounts, the evals, a decision log, a person who has already changed something alone, and a model-update procedure.
Imagine hiring a master builder to remodel your kitchen. If he is the only one who knows where the pipes run and where the shutoff valve is, you did not remodel a kitchen. You rented his memory. A deployment engineer for AI is that builder. Your job is to ask for the blueprint and the valve key before you pay the last installment.
AI-generated summary
Anthropic announced Claude Frontier Academy on Friday, October 2. The pitch: $100 million to train 10,000 Frontier Deployed Engineers by the end of 2027. Cohorts are already running in San Francisco, New York and London, with engineers from Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk.
A forward deployed engineer (FDE) sits inside your company until the AI works in the real process. Palantir made the model famous. Now the lab that builds the model is paying to multiply it.
That same day Mark Ajzenstadt wrote the sharpest take I saw. Many companies, he said, will rent this capability before they can build it. “The question is what stays when the engineers leave: working software, documented decisions, and a team that can keep improving it.” His rule: “Rent it from someone who plans to hand it over.”
I agree. Here is how I would turn that into something you can put in a contract.
Anthropic’s own curriculum teaches handover
Read how the training works. A few days in person, running a simulated deployment “from the right use case through security review to handover.” Then a 12-week residency where the engineer leads a real Claude use case at their own organization. Everyone arrives with a named project. First credentials are expected in early 2027, and entry is by nomination only.
Look at who is on the list. Accenture, Bain and Deloitte sell deployment. Commonwealth Bank, Morgan Stanley and Novo Nordisk are training their own people. Those banks and that drugmaker chose to grow the FDE at home. A mid-size company in Central America is unlikely to nominate anyone to San Francisco. Most likely it rents.
When you rent, handover is not a closing detail. It is the product you are buying.
Five things that must stay
Imagine the engineer vanishes tomorrow. What would you miss?
Your accounts, first. Repository, cloud project, API keys, the Claude workspace: all in your company’s name from day one. If it starts in the consultant’s account and “migrates at the end,” the migration does not happen.
Second, the evals. An eval is a set of real cases with the correct answer already known. It runs every time something changes and tells you whether the system got worse. I wrote about why evals are what stays when the model changes. If the consultant built them without your people in the room, you do not own them. You rent them.
Third, a decision log. One line per rule: why the agent cannot approve payments above a certain amount, why a human reviews step three. In a year nobody remembers the conversations. The log remembers.
Fourth, one person on your side who has already changed something alone. To me this is the real test. Having attended the meetings does not count.
Fifth, the update procedure. Models change every few months. Someone has to know what to rerun when the next one lands.
Writing it into the contract
Turn the list into payment terms.
The most useful one is a handover test. Hold back the last part of the invoice until someone on your team changes a rule, runs the evals and ships the change while the consultant watches with hands off the keyboard. If they cannot, the project is not done.
Ask for documentation weekly, not at close. What gets written on the last day gets written badly, or never. And require that someone from your side work next to the engineer from week one, with hours blocked in their calendar rather than “when there is time.”
One disclosure about the Academy itself: its credentials are Claude credentials, and IQ Source works with Claude, so I have a bias. Still, the more your system depends on one vendor, the more your evals matter, because they are what lets you switch models without starting over.
How this fits with what I already wrote
On October 11, I covered Aaron Levie’s list of seven deployment jobs, and the conclusion was that three decisions cannot be outsourced. That post is about what you decide before hiring. This one is about what you hold when the contract ends. Levie wrote on October 4 that FDEs and new deployment firms are jobs AI is already creating. More supply of people to rent from. It does not mean they all hand over.
And the trade itself is old. In the consultant-to-FDE rebrand I argued it is the third name for the same job since 1990. The label changes. What you have to negotiate, the ending, does not.
About to sign a deployment contract and want someone to read the handover clause before your lawyer does? Send it to us and we will go through it with you
Frequently Asked Questions
Claude Frontier Academy is the program Anthropic announced on October 2, 2026, backed by $100 million to train 10,000 Frontier Deployed Engineers by the end of 2027. It combines several days in person with a 12-week residency inside the participant's own company, where they lead a real project built on Claude.
A forward deployed engineer, or FDE, works inside the client company to learn how it operates and build software that fits that work. Palantir popularized the model. Companies want FDEs to take AI agents from prototype to production, because that takes knowing the business, not only the technology.
When an AI deployment ends, your company should hold the accounts and repositories in its own name, the evals that measure whether the system works, a log of the decisions made, one employee who has already changed the system without vendor help, and the procedure for updating it when the model changes.
Make the handover a payment condition. Release the final part of the invoice when someone on your team changes a rule in the system, runs the evals and ships the change without the consultant touching anything. Also require documentation every week instead of at the end, and keep everything in your accounts from day one.
Related Articles
Aaron Levie's list: the real work of deploying AI agents
Aaron Levie listed seven jobs behind putting AI agents into a company. Three of them are yours to own before you pay anyone to do the deployment for you.
Clef arrived two weeks after Jev. Design for the swap
Cloudflare released Clef, an open decision model compatible with Jev's API. Neither wins everywhere, so what matters is how cheap it is to change your mind.
Karp Attacked AI Pricing on CNBC. A Bill Proved Him Right.
Alex Karp accused OpenAI and Anthropic of extracting value through token pricing. Days later Pylon's CEO watched his bill jump from $400K to $1.4M.