Why Enterprises Are Racing to Build Their Own Forward-Deployed AI Engineers

The scarcest resource in enterprise AI this fall is not GPUs or frontier models. It is the engineer who can take a model into a messy business process and come out the other side with a production system. This week, Anthropic put $100 million behind training those people inside its customers, and Barclays described how far it has already pushed AI into its engineering and operations. Together they point to a shift in how large companies staff AI: less reliance on vendors' own field engineers, more effort to grow that skill in-house.

What is a forward-deployed engineer?

The term comes from software companies that embed their own engineers with customers to make a product work in a specific environment. In AI, the job means picking a use case, wiring a model into a company's systems and data, getting through security review, and handing over something the business can run.

Anthropic calls its version the Frontier Deployed Engineer. On October 2, it announced Claude Frontier Academy, a $100 million program to train 10,000 of these engineers by the end of 2027. The company says people with the skills to make AI work inside a business have become the hardest talent to find, and that a small group of skilled people drives an outsized share of AI's value in enterprises. It offers no statistics for either claim.

Box CEO Aaron Levie made a similar observation the day before. Enterprises are creating their own forward-deployed engineer roles to connect models to business workflows, he wrote, according to a summary by The Neuron, because knowledge of a company's systems and processes still matters.

How does Anthropic's program work?

Anthropic modeled its first program, a residency, on medical training. Engineers are nominated by their organizations and arrive with a named Claude project to lead. They start with a multi-day, in-person course with Anthropic engineers that walks through a simulated enterprise deployment, from choosing a use case to security review and handover, followed by a graded practical exercise on a new scenario.

Those who pass enter a 12-week residency, leading a real project at their own company with support from Anthropic engineers and their cohort. A final assessment awards a Frontier Deployed Engineer badge, with the first expected in early 2027. Cohorts are running in San Francisco, New York and London. Prior experience building AI agents is not required, but candidates are expected to be hands-on software engineers who have built with large language models.

The launch partners mix consultancies and end customers: Accenture, Bain, Capgemini, Deloitte and McKinsey alongside Commonwealth Bank of Australia, Morgan Stanley and Novo Nordisk. The program builds on Anthropic's partner network, which the company says spans 46,000 firms and has issued more than 175,000 Claude certifications. Anthropic's page does not mention what participants pay.

Steve Corfield, Anthropic's global head of business development and partnerships, framed the pitch around depth, saying no AI company has invested in this talent inside its customers "at this depth." The company did not say how it will measure success beyond the 10,000 target.

What does this look like inside a bank?

Barclays offered a view of the scale these engineers are meant to support. In an Anthropic customer announcement on October 1, the bank said it expects Claude Code to reach half of its developers by the end of 2026 and a majority of its software engineers in 2027.

The bank already runs Claude in daily operations. A Colleague Knowledge Assistant, live since 2025 and built on retrieval-augmented generation, helps Barclays UK staff find answers while serving more than 20 million retail customers. More than 16,000 employees have adopted it, and it has handled over a million searches. In Global Markets, Claude classifies and routes about 120,000 incoming client emails a day. Barclays says it applies governance, security controls and human oversight to these uses.

Commonwealth Bank of Australia, one of the Academy's partners, gave a productivity data point of its own on the Academy page: Group CTO Rodrigo Castillo said teams using AI tools produced up to three times more code changes over the past year. Like most figures in vendor announcements, these are self-reported and do not show the cost or quality side of the ledger.

Why can't enterprises just rely on vendors?

Vendors have every reason to train customer engineers. It speeds adoption of their own models, and Anthropic's program is built around Claude projects. Enterprises should weigh that. A team certified on one vendor's tools may default to that vendor, even as competing models arrive every week.

But the underlying need is real, and AI Connect event speakers have described it for some time. Adam Malone, who runs forward deployed engineering at PromptQL, described a ladder of AI adoption in a AI Connect webinar. At the bottom is AI theater, where a CEO demos something at an all-hands. "Everyone claps. Nothing changes," he said. The harder work, in his account, is capturing the business knowledge that lives in people's heads, which outside engineers rarely have.

Training also has to fit the job. "Meaningful upskilling really comes from when it's contextual and not generic," Arpit Gularti of The Clorox Company said at COLLIDE 2025, in a panel on AI change management and upskilling. A residency built around a real internal project fits that description better than a generic certification course.

Experience at AI Connect events also suggests that organization matters as much as individual skill. Leaders from Snowflake, Revefi and The Coca-Cola Company argued in a webinar that most AI pilots stall for organizational reasons, not technical ones. A company can train strong engineers and still stall if no one owns the business process they are changing.

What should enterprise leaders do?

• Define the role before you fund it. Decide whether your deployment engineers sit in IT, a central AI team or the business units, and who sets their priorities.
• Pick people with domain knowledge. The most valuable candidates often know the business processes, not just the models.
• Keep training vendor-neutral where you can. Use vendor programs for depth, but make sure evaluation, security review and handover practices work across providers.
• Measure results, not badges. Track projects shipped to production and their business outcomes, as Capgemini's Geoffroy Pajot noted on the Academy page when he said clients care about solving real problems, not certification counts.

Anthropic's first badges arrive in early 2027. By then, the companies that gain most will likely be the ones that treated the program as one input into their own AI engineering bench, not a substitute for building one.

Frequently asked questions

What is a forward-deployed engineer in AI?

An engineer who takes an AI model into a specific business, picks a use case, connects the model to company systems and data, gets it through security review and hands over a production system.

What is Claude Frontier Academy?

A $100 million Anthropic program announced October 2, 2026, to train 10,000 Frontier Deployed Engineers by the end of 2027 through a course, a 12-week residency and assessments.

Who can join Claude Frontier Academy?

Participation is by nomination. Candidates are hands-on software engineers who have built with large language models and arrive with a named Claude project to lead.

Disclosure: AI Enterprise Magazine uses AI tools in its editorial process, including Anthropic's Claude models. Anthropic had no role in this article.

Filed under: People, Culture & Leadership, AI Strategy & ROI, Agentic AI