How AWS, OpenAI and Anthropic run forward deployed engineering

Four public programmes exist. This page sets out what each company says about its own, with a note on why self-declarations need reading carefully.

A warning before the detail. Everything below comes from the companies themselves: newsroom posts, partner blogs, documentation, job adverts. A company describing its own delivery model is telling you what it intends, not what a customer received. We have no independent verification of any of it, and neither does anyone else writing about this. Read it as intent.

Palantir: the original

Palantir created the role and has used it since the mid-2000s, embedding engineers with one customer at a time to build on Foundry or Gotham against that customer's real data. The Palantir model covers the origin and why the original condition still governs.

The most telling recent thing Palantir published is not about people. It is a product: AI FDE, documented as an interactive agent that operates Foundry through conversational commands, building pipelines, creating and updating ontologies, writing and testing functions and building applications, in a closed loop, inside the permissions the user already holds.

A company naming a product after its own signature role is making a statement about which half of that role was mechanical. Is the role being automated takes that apart.

AWS: the largest commitment, and the clearest one

In June 2026 AWS announced a Forward Deployed Engineering organization backed by a $1 billion investment, embedding thousands of engineers with customers to co-develop and deploy production AI systems.

Three things in that announcement are worth more than the dollar figure.

Pricing. AWS said deployments are structured around shared goals and business results rather than billable hours. That is a commitment a buyer can test in a contract.

Self-sufficiency. It named customers being self-sufficient once an engagement ends as a design goal. That is the single most important sentence any vendor in this space has published, because the failure mode of the model is the engagement that never ends.

Positioning. Francessca Vasquez, VP of Frontier AI Engineering and Services, framed the demand as customers having moved past exploring what AI can do and wanting it central to how they operate. AWS also named regulated sectors, financial services and government as target areas, which matches where the hiring data says the demand is.

AWS again: the partner programme

A month later AWS extended the model to consulting firms, in Forward Deployed Engineering for Partners, launched 30 June 2026 with a limited set of partners.

The requirements are the interesting part. Partners must build ring-fenced, AWS-credentialed engineering teams that pass an AWS-defined technical bar, described as a production-engineering standard validated against the same methodology AWS's own forward deployed teams use, before engaging a customer. They must own reusable delivery harnesses including domain ontologies, evaluation frameworks and capability registries.

The programme runs in three phases: AWS engineers embedded with partner teams, then AWS support shifting toward helping the partner scale its own practice, then partner teams working independently.

It also names a methodology, Agentic Process Transformation, which directs customers to start with a business process, reimagine it for an agentic world, deploy it to production, and demonstrate measurable results.

Read the credentialing requirement as an admission worth having: putting embedded engineering inside a consulting firm does not happen by relabelling the delivery team. It needs a separate team and a separate bar. FDE vs systems integrator covers what that means for a buyer.

OpenAI

OpenAI posts both Forward Deployed Engineer and Forward Deployed Software Engineer titles across San Francisco, New York, Seattle and London.

What it states about its own roles: up to 50 percent travel is required, and the work involves embedding with strategic customers to understand their business and technical requirements.

On scale, a16z's Trading Margin for Moat recorded 22 open forward deployed and solutions engineering roles out of 311 total OpenAI positions when it was published on 4 June 2025. That is a third-party count of public postings rather than a company statement, which makes it slightly better evidence than a self-description.

We could not verify a compensation band for the role. Levels.fyi does not have enough submissions to show a level breakdown for OpenAI's forward deployed engineer title, and the confident ranges in circulation do not trace to a primary source.

Anthropic

Anthropic posts a Forward Deployed Engineer role on its Applied AI team, embedding with strategic customers. Its own posting describes delivering technical artifacts such as MCP servers and agents for production use, providing deployment support in enterprise environments, and identifying repeatable deployment patterns to feed back to product and engineering.

That last item is the productization loop stated as a job duty, which is the right place for it. The productization loop covers why a function without it drifts into a services business.

What to take from all four

The two commitments to demand. Outcome-based pricing rather than billable hours, and customer self-sufficiency at a named date. AWS published both. Ask every vendor for both, in the contract, and treat reluctance as the answer.

The feedback loop as a duty, not an aspiration. Anthropic's posting names identifying repeatable patterns as part of the job. If a vendor cannot say who owns that and on what cadence, the custom work will not generalise.

Credentialing over relabelling. AWS's partner bar exists because the default consulting skill set is different. Ask what bar a vendor's engineers passed, and who set it.

And the limit of all of it. These are self-declarations. The only ones that mean anything are the ones a customer can enforce, which is why the two dates in our contract page matter more than any announcement.