Forward Deployed Engineering Trends 2027: The FDE Talent War and the Mid-Market Gap

Key Takeaways

• The FDE talent war is real, and the big buyers are winning it. Postings for the title grew 1,165% year over year in 2025, and the AI labs are now buying whole consulting firms to staff their own deployment arms.
• FDE has become a product line. In May 2026, OpenAI and Anthropic each launched deployment businesses backed by private equity money. That changes who the talent works for.
• The mid-market is where the gap shows. Nearly every mid-size company now uses generative AI, and most of them say they need outside help to make it work.
• Hiring your way out is the slow path for a portco. A small team cannot outbid the labs for a scarce role, and one FDE without a team around them rarely ships much.
• The partner model wins in 2027, with conditions. Rent the engineering, own the knowledge, and judge partners on what is in production.

Two of the most valuable AI companies on the planet did something in May 2026 that would have sounded odd two years ago. Within a week of each other, they started consulting firms.

On May 5, Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced a new enterprise services firm to bring Claude into the core operations of mid-size companies. Goldman's Marc Nachmann framed it as giving the mid-market access to forward deployed engineering talent that has mostly been reserved for the largest enterprises. Six days later, OpenAI launched the OpenAI Deployment Company with more than $4 billion in initial capital led by TPG, and bought the consultancy Tomoro to get about 150 forward deployed engineers on day one.

When the model makers start buying services firms, something about the market has changed. My read: the scarcest input in enterprise AI right now is the person who can sit inside a business, figure out where the work actually happens, and ship something that runs in production. There aren't enough of those people.

That is the FDE talent war. The biggest players are hiring, acquiring and funding their way to as much of that talent as they can get. Which leaves a real question for everyone else, and especially for the mid-market companies and PE portfolio companies we work with. Where does your deployment capacity come from in 2027?

The FDE Title Grew Faster Than the People Who Can Do the Job

Forward deployed engineer, a title that now appears on roughly every job post with a customer in it, has had a wild couple of years.

Bloomberry's analysis of 1,000 FDE job postings found postings for the title grew 1,165% from January through October 2025 compared with the same months of 2024. Fast Company, citing the Financial Times, put the growth at more than 800% over the first nine months of 2025. Different windows, different methods, same direction.

The supply side did not move like that. You can't produce an engineer who writes production code, talks to an operations lead, and understands a messy ERP in a year. The same Bloomberry data shows that 60% of FDE postings that list experience ask for three to five years, and 68% of postings require travel. That is a narrow slice of the engineering market, and it is the slice everyone wants.

So pay goes up. Bloomberry puts the median advertised base salary at $173,816. A recruiting firm's 2026 write-up puts total compensation at AI labs at $350,000 to $550,000 for mid to senior FDEs. Read that second number with the source in mind (recruiters benefit from a hot market). The direction still matches what anyone hiring in this space feels.

Separate the title from the work

When you read "FDE" on a resume or a vendor's slide, ask what the person shipped and where it runs today. The title spread faster than the skill did. A demand curve like that pulls in plenty of people whose job changed names and nothing else.

The Labs Turned Deployment Into a Business Line

The part of 2025 and 2026 I think a lot of people missed: forward deployed engineering started as a way to sell software. It is turning into the product.

The progression is easy to trace. In mid-2025, The Decoder, citing The Information, reported that OpenAI's FDE-staffed consulting work started at $10 million per client. That was a service for very large companies with very large budgets. A year later, both leading labs had stood up separate deployment companies, funded by private equity, with their own hiring plans and acquisition budgets.

I think this is mostly good news. The labs know their models better than anyone, and having their engineers close to real deployments will make the models better at real work. The AltsWire report says Anthropic engineering and partnership staff will be embedded in the new firm's team for exactly that reason.

But it also concentrates the talent. Every FDE who joins a lab-backed deployment company is an FDE who isn't available to hire, and the economics of those firms will pull them toward the biggest accounts they can serve. That is how services businesses work. Large accounts pay for the overhead.

Expect the big ventures to start with their own networks

Both ventures come with built-in distribution. AltsWire notes the Anthropic firm also has backing from a consortium including General Atlantic, Leonard Green, Apollo, GIC and Sequoia, which gives it access to hundreds of portfolio companies. If your company sits inside one of those portfolios, you may get a call. If it doesn't, plan as if you won't.

The Mid-Market Is Where Demand and Supply Split Apart

The firms with the most to gain from an embedded engineer are often the ones least able to hire one.

RSM's 2025 Middle Market AI Survey found that 91% of middle market firms use generative AI, up from 77% a year earlier. Seventy percent of those users said they need outside support. Among firms that felt unprepared, a lack of in-house expertise was the top reason, at 39%.

Put that next to the hiring data. The roles exist, they are expensive, and Bloomberry found that 58% of them sit at companies with 11 to 200 employees. That profile looks like AI startups hiring FDEs to deploy their own product at their own customers. The talent flows toward vendors and labs. The buyers, the manufacturer or the regional services firm with 800 people and a thin IT team, are left trying to recruit from the same pool with less money and a less exciting pitch.

That is the gap. Demand for deployment work is spreading out to the mid-market, and supply is concentrating at the top.

Count the team you'd need, not the hire

A single FDE inside a mid-size company is usually a lonely job. Production AI needs someone on data, someone on the application, someone on evaluation, and someone who owns it after launch. Before you open a req, list every role the first workflow will need. If the list is longer than your budget, that tells you the answer.

Hiring an FDE Team Is the Slow Path for Most Portcos

I'll say it plainly. For most PE-backed mid-market companies, building an in-house FDE team in 2027 is the wrong first move.

That is a strong claim, so let me complicate it right away. There are companies where it is right. If AI is core to the product, if you have a technical leader who can manage these people, and if you can offer equity that competes, hire. Long term, the companies that own their AI capability will have an edge.

For everyone else, the math is rough. A hold period might be four or five years. A full search for a role like this can take months. Then the new hire needs time to learn the business, earn trust with the operators, and get access to systems. Meanwhile the operating plan assumed the AI value showed up this year.

Here's how I'd lay out the options a mid-market company actually has.

Decide what you need to own

The real question is which parts you need to own and which parts you can rent. Own the data, the workflows, the evaluation criteria and the decision about what gets built. Rent the engineering hours until you know which capabilities deserve a permanent team.

The Partner Model Wins in 2027, If It Leaves Something Behind

My prediction for 2027 is that most mid-market FDE work will run through partners. Some of those partners will be the lab-backed firms. Many will be independent shops that can move faster on smaller programs. A lot of them will be firms that renamed their staff augmentation team last year.

So the model alone won't save you. What makes a partner model work is what it leaves behind.

A good embedded engagement looks a lot like having your own FDE team for a while. The engineers join standups, work in your repos, follow your release process and demo every sprint. Code, documentation, evaluation harnesses and runbooks stay with you. Your people pair with them the whole way, so the knowledge transfers as the work happens.

That is how we run PressW's Forward Deployed AI Engineering: senior engineers embedded as a pod that can flex from one person to a full squad across AI, data and application work. One example: for a venture capital firm, our team built a data room assistant that saves an estimated 60+ hours a month during LP due diligence, integrated directly with the firm's data room provider.

Judge every partner on what's in production

Three questions sort partners fast. What did your last three engagements put into production, and are they still running? Who on my team will be able to maintain this after you leave? What does your evaluation harness look like, and who owns it at the end? If the answers are vague, the FDE label is doing a lot of work.

PE Firms Will Treat Deployment Capacity Like Shared Infrastructure

One more trend I'd bet on. Funds will stop leaving AI deployment to each portfolio company and start handling it at the fund level.

The lab ventures already point this way. Their pitch to investors leans heavily on portfolio access. That logic works for any fund with ten or more operating companies. Most of those companies face the same few workflows: finance close, customer service, quoting, document review. Building the playbook once and deploying it across the portfolio beats paying for ten separate learning curves.

The open question is how far that goes. Do funds hire their own central AI engineering teams? Contract one partner across the portfolio? Mix the two? I don't think anyone has a clean answer yet, and I'd be skeptical of anyone who claims one. What I'm fairly sure of is that the funds that treat deployment capacity as shared infrastructure will see results faster than the ones that leave each CEO to figure it out alone.

Start with a portfolio scan

Before you staff anything, find out where each company stands on data, systems and people. An AI and data readiness assessment across the portfolio shows which companies can use an embedded team right away and which need data work first.

Where This Leaves Mid-Market Operators

The FDE talent war isn't going to cool off in 2027. The labs have money, the role is scarce, and the work is valuable. That combination keeps prices high and talent concentrated at the top of the market.

I'm optimistic anyway. The labs moving into deployment validates something we've believed for a while: getting AI into production is an engineering problem inside a specific business, and it takes people who show up and do it. The mid-market just needs a way to get those people without waiting in line behind the Fortune 100.

I don't know exactly how the partner market shakes out. Some lab-backed firms will reach further down-market than people expect. Some independents will get acquired. What I'd tell any operator right now is to pick one workflow that matters, get an embedded team on it, and insist that your people own it when the team steps back. If you want to compare notes on where your company or portfolio stands, Book a Discovery Call.

FAQ