Who supplies forward deployed engineers and when to build your own team

The labs' own FDEs work with their most strategic customers. OpenAI and Anthropic say so in their job postings. Most companies reach this model another way: through a lab's deployment company, an integrator or a small team. Below I describe what each kind of supplier gives you, when your own team pays off and what to write into the contract. I read the sources on 3 November 2026, and they are listed at the end.

AI labs and their deployment companies

OpenAI and Anthropic hire FDEs into their own teams. OpenAI's San Francisco posting has its FDEs lead deployments "alongside our most strategic customers". Anthropic's London posting says the same of the FDEs in its Applied AI team. A mid-sized company should not assume it will get an engineer straight from a lab.

Both labs have also set up separate services firms with investors. Their announcements say these firms extend the labs' delivery capacity. OpenAI launched the OpenAI Deployment Company, DeployCo for short, on 11 May 2026. It started with more than $4 billion of investment, and OpenAI holds the majority and control. Its job postings say DeployCo works with many of the world's largest companies. A typical engagement begins with a diagnostic of where AI can add the most, and then the team picks a few workflows with the customer and builds them in the customer's systems on OpenAI models.

On 4 May 2026 Anthropic announced a new services firm for mid-sized companies. It made the announcement together with Blackstone, Hellman & Friedman and Goldman Sachs. Its examples were community banks, mid-sized manufacturers and regional health systems. On 15 July the firm launched as Ode with Anthropic. Its core is the team of Fractional AI, acquired in May, together with engineers from Anthropic. On its website Ode says it does not hand off a repo and disappear. It works with the customer's team from roadmap through deployment.

This kind of supplier gives you the shortest path to people who know the model from the inside. You pay for that by choosing your model provider when you sign. DeployCo connects OpenAI models to the customer's data, and Ode builds on Claude. The background to this shift, with the postings and the numbers, is in my piece on why AI labs now build whole companies around this work.

Platform vendors: FDEs who deploy their own product

Palantir's current posting says the company pioneered the position. Its Forward Deployed Software Engineer works at the customer but builds on Palantir's platforms.

Other software vendors now have similar teams. Salesforce writes that its FDEs bring its most advanced AI solutions to life for its biggest clients. The platform in question is Agentforce, its agent platform. ServiceNow has an FDE team of its own. Together with Accenture's FDEs it builds agent workflows on the ServiceNow AI Platform, inside the customer's environment.

An FDE like this knows the product best and will get it running fast. The job is to deploy that platform, because the choice of tool was made earlier, when the licence was bought. This route makes sense when you already have the platform. Ask what happens to the code and configuration if you ever leave it.

Integrators: FDE practices at Accenture, Infosys and TCS

Large integrators announced FDE practices through 2026. On 18 March Accenture launched an FDE practice with Microsoft. The two companies promise "thousands of AI-skilled engineers". Microsoft provides the platform, and Accenture leads process change and deployment across many countries. On 6 May Accenture and ServiceNow announced a similar program. Each engagement there gets a pod built around the value chain of that customer.

The Indian integrators use names of their own. Infosys CEO Salil Parekh said on 23 July 2026: "Our plan is to have 6,000 Frontier Engineers over the next few years." He gave no current headcount. Asked about FDEs, he answered about Frontier Engineers and repeated the objective of 6,000. On 9 July TCS COO Aarthi Subramanian described its rapid build engineers as a role "very similar" to an FDE. Its FDEs work in teams called pods, or squads. She gave no count of the FDEs TCS has today. The target is at least 1% of employees working in the new operating model. She added that it is a transition and that the definition of an FDE is still evolving.

The labs work with integrators too. In February 2026 OpenAI announced its Frontier Alliances with BCG, McKinsey, Accenture and Capgemini. The partners work alongside OpenAI's FDE team. Anthropic writes that systems integrators in the Claude Partner Network lead the deployments at the largest enterprises. It names Accenture, Deloitte and PwC. Before you choose a partner, check what official Claude partner status means and which certifications count toward its tier.

An integrator gives you scale, offices in many countries and people who manage change. These practices are new, though. Before you sign, ask for the names of the people who will join the team, and ask which of them actually write production code.

Boutiques, independent engineers and collectives

The smallest suppliers are small deployment firms, independent engineers and collectives of freelancers. The labs' deployment companies turned to firms like these. In May OpenAI agreed to acquire Tomoro, an applied AI consulting and engineering firm. The deal is meant to bring about 150 FDEs and Deployment Specialists to DeployCo. Ode grew out of Fractional AI, an applied AI services firm.

A small team can start quickly and often works on the model you choose. The risk lies in scale. When one person leaves, part of the knowledge leaves with them. That is why knowledge transfer weighs more with a boutique than with a large integrator.

I work in this segment myself, so judge me by the same questions. With a collective of experienced specialists we join the customer's team or take the whole project, from architecture to production. I describe both options on the page about team augmentation and projects delivered end to end.

Recruiters: when you want a permanent FDE hire

Recruitment agencies find candidates for a permanent role or a contract. They do not deliver a deployment. They supply candidates, and the brief, the work sample and the acceptance stay on your side. I do not recommend specific agencies here. I found no source that allows a fair comparison.

A title on a CV does not tell you what the candidate did. Check whether they took a system into production at a customer and whether they ran it afterwards. Before you pay for a search, prepare a one-page brief and a paid work sample that lets you compare candidates.

Your own team or a supplier: six criteria

Decide separately for each process. The table below is my proposed set of criteria. For each row, mark the column that better describes your situation.

When you build a team, you compete for the same people as the labs and AI startups. OpenAI's San Francisco FDE posting prints $185,000 to $300,000 a year plus equity. DeployCo's US posting prints $170,000 to $400,000 a year. A posting from fonio, a company that builds AI phone agents, prints €60,000 to €85,000 a year for Warsaw. These are single postings read on 3 November 2026, not market rates. I collected more salary ranges that employers print in their own FDE postings in a separate guide.

Hiring takes time as well. Until the day the new person starts, the process does not change. Put that time into the schedule before you compare costs. Consider a mixed setup too. A supplier builds the first process, and your engineer works in its team from day one and takes the system over after acceptance.

What to write into a contract with an FDE supplier

Three clauses protect you most: ownership, handover and acceptance. Agree the wording with your lawyer. Below is what I think they should contain.

Ownership. Code, prompts, test sets, configurations and test data go into your repository from day one. The same applies to the accounts with the model provider and to the API keys. Also write in the names of the people on the team and the rule for replacing them.

Handover. Describe what you receive at the end: setup instructions, an incident procedure and a list of known limits. Before acceptance, your own person starts the system from the instructions and restores the previous version. Also agree how long the supplier answers questions after the work ends.

Acceptance. Agree the test case set and the quality thresholds before the build starts. Add failure cases: missing data, an API outage and a request outside the scope. Compare the result with the current process.

Add two items that are easy to miss: who pays for model calls after launch and who maintains the system. Before you sign, it helps to see a quote split into build costs and the costs that grow after launch.

Before you send your first request, write the process, the owner after launch and the acceptance criteria on one page. Send the same page to two suppliers of different kinds and compare the answers.

Frequently asked questions