Why AI companies hire Forward Deployed Engineers: the a16z thesis, 15 months later

In June 2025 Andreessen Horowitz published the argument that explains why the Forward Deployed Engineer job exists at all. Fifteen months later I went and checked what happened to it, with the same instrument they used. This is what I found, and what it means whether you are trying to become an FDE or trying to hire one.

The thesis, in one sentence

Joe Schmidt, a partner at a16z, wrote it on 4 June 2025 under the title "Trading Margin for Moat". The sentence everyone quotes: enterprises buying AI are like your grandma getting an iPhone. They want to use it, but they need you to set it up.

The argument underneath was uncomfortable for a Silicon Valley that had spent a decade worshipping product-led growth. Slack, Figma, Notion and Dropbox win by being adopted from the bottom up with a credit card. But during a platform shift, the companies that end up owning the market are the ones that do heavy implementation work and eat the margin hit: Salesforce, ServiceNow and Workday all sold platforms that needed configuration, integration and services before they did anything useful.

The numbers he used are historical, not opinion. At IPO, ServiceNow's gross margin was 63.2 percent and Workday's was 54.1 percent, both far below the 80 percent that software investors treat as the ideal. Salesforce burned over $52 million to generate $22 million of revenue before it had a partner ecosystem. By 2024, ServiceNow was back to 79 percent and Workday to 75.

The margin comes back. The moat does not go away.

Why that produces a job title

Enterprise AI has an even bigger implementation requirement than cloud software did, because an agent is only useful once it is connected to the company's real systems, real data and real permissions, and given the context to act on them. Somebody has to sit inside the customer's environment and do that. a16z's phrase for it is that agent onboarding looks like employee onboarding: active management, guided learning, and read and write access to internal systems. For a human employee that job is done by a manager. For an AI agent it is done by a forward deployed engineer.

That is the whole role. Not a solutions engineer who demos and hands off, and not a product engineer who never meets a customer. An engineer who writes production code inside someone else's stack, holds the room with the customer's executives, and ships the first use case into production.

The evidence a16z used, and the same measurement today

To show the labs were already staffing this, the post counted OpenAI's careers page. In June 2025, 22 of its 311 open roles were forward deployed or solutions engineering.

I read the same board on 3 September 2026 through OpenAI's public posting API, with a fabricated company slug as a control to prove the instrument returns 404 when nothing is there. Today the board carries 770 open roles, and 18 of them have "Forward Deployed" in the title.

The count is not the interesting part. The structure is:

• Two of the eighteen are managers, including a Manager of Forward Deployed Engineering.
• The function has verticalised: healthcare roles in New York, San Francisco and Seattle, legal roles in New York and San Francisco.
• There are government and security variants, both in Washington DC.
• The roles sit in eight cities across four continents: SF, Seattle, NYC, DC, Seoul, Singapore, Sydney and Tokyo.

In June 2025 a16z counted a category. Today it is an org chart with managers and verticals. That is what a function looks like once it stops being an experiment.

I am deliberately not drawing a trend line from their 22 of 311 to my 18 of 770. Their definition included solutions engineering under a rule I cannot reproduce, and a loose matcher over today's board returns 24 roles including three in technical accounting, which is the matcher lying, not the market. The honest comparison is structural, and the structural story is stronger anyway.

What happened in the fifteen months in between

The thesis was not a prediction that aged badly. Three dated data points since:

• On 30 August 2026 The New York Times covered the role, citing Indeed data showing FDE listings up 730 percent year over year.
• On 2 September 2026 Wonderful, an enterprise AI company founded in early 2025, raised $550 million at a $5 billion valuation and said in its own announcement that it will hire "hundreds of cracked engineers, FDEs, and former founders". Its careers page lists 32 cities, four of them in Latin America and none in the United States.
• Every one of the recruiting firms that now rank for "hire forward deployed engineers" describes the same two-axis screen: an integration exercise plus a customer scenario. The role has a standard interview shape now, which is another sign it is permanent.

What this means if you want the job

The buyers of your labour have decided the role is permanent, and they have decided what it tests. Not LeetCode. A deployment case: the questions you ask before you build, how you scope a first use case that can reach production in weeks, tool-use design, orchestration you own rather than a framework hiding it, evals that catch the failure that matters, a rollout with a canary and a rollback, and one more that a16z and the recruiting firms both name and that most candidates never think about.

That seventh one is the product feedback loop. a16z's seventh best practice for services teams is that forward deployed engineers are the people closest to the customer, and the job includes running what they learn back into the product without games of telephone. A competitor's page on FDE recruiting puts it more bluntly: the best FDEs write product feedback, not tickets, and the instinct to push a customer-specific hack back into the core product is what separates an FDE from a contractor doing bespoke integration work.

We added that dimension to our own grading rubric this week, so the free timed case now scores all seven. Sit the case. It takes an hour, and the graded review names the dimension that cost you the points.

What this means if you are hiring one

You are hiring into the market a16z described, fifteen months in. The title is two years old, so almost nobody has it on a CV, and the applicant flood the NYT described means a resume screen cannot separate the people who have deployed from the people who have read about agents. The only thing that can is a case, graded against the seven things the job actually requires.

That is what we do. Here is how it works, and here is a full sample graded report for the healthcare case, so you can judge the instrument before you judge a candidate.

Sources: a16z, "Trading Margin for Moat: Why the Forward Deployed Engineer Is the Hottest Job in Startups", Joe Schmidt, 4 June 2025. OpenAI role counts read from the company's public Ashby posting API on 3 September 2026. The New York Times, 30 August 2026. Wonderful's own funding announcement and careers page, read 2 and 3 September 2026.