Skip to content
Nolli
Request a demo
Nolli Learn

Guide · See

What is a forward-deployed engineer?

What forward-deployed engineers do, the playbook they follow to put AI where it belongs, why they are hard to scale, and how to give every person on your team the same capability.

The short answer

A forward-deployed engineer (FDE) is an engineer who embeds with a customer’s team, learns how the work actually gets done, and then builds and deploys software inside that team’s real workflows. Palantir pioneered the role in the early 2010s. Today it is how many companies get AI agents into production.

By Nadi Haque, Founder, NolliUpdated 9 min read

Key takeaways

  1. FDEs start with the work, not the technology. They map how the process really runs before they build anything.
  2. They learn from three sources: the people who do the work, the systems of record, and the documents.
  3. FDEs work, but they are scarce, expensive and temporary. The next step is giving every person FDE capabilities.

What does a forward-deployed engineer do?

A forward-deployed engineer works inside a customer’s business rather than on a product team. They sit with the people who do the work, find where the process really breaks, and ship software that fixes it. Then they stay until it runs in production.

Most FDE teams follow some version of the same playbook:

  1. Embed with the teamSit with the people who do the work.
  2. Map the real workHow it actually runs, not what the SOP says.
  3. Decide where AI belongsStep by step: delete, code, agent or person.
  4. Build inside existing toolsAgents that work where people already work.
  5. Prove it, then improve itMeasure against the old way, and keep tuning.
Figure 1. The FDE playbook. The order matters: the work is mapped before anything is built.Source: Varick Agents, AI forward deployed engineering

Varick Agents, an FDE firm, spends at least four weeks watching how the work is really done before it touches a model, according to Daniel Kornum of Varick. The firm’s FDE guide gives the reason in one line: “the documented process is rarely the real process.” Skipping that step is expensive. When a team builds for the documented process, Kornum writes, the automation handles roughly 70% of the volume and breaks on the other 30%, which can create more work than it saves.

Where did the FDE role come from?

Palantir created the role in the early 2010s. Its forward-deployed software engineers, known internally as Deltas, were placed with customers such as government agencies, where a failed deployment had real consequences. According to Varick Agents, Palantir engineers worked alongside special forces in Afghanistan around 2010, shipping fixes overnight based on what operators reported each day.

The model spread with AI. In 2025, Andreessen Horowitz called the forward-deployed engineer the hottest job in startups, arguing that AI applications only deliver once someone wires them into a company’s systems and teaches them its context. Even model providers such as OpenAI and Anthropic hired forward-deployed engineers to win enterprise customers.

How do FDEs find where AI belongs?

Good FDEs never trust a single source. They combine three:

People

Interviews with everyone who touches the work. Most know-how is never written down: who really decides, which steps are for show, how exceptions get handled.

  • Accounts payable
  • Billing
  • FP&A
  • Deal desk
Systems of record

A few weeks of CRM or ERP data shows what actually happens, how often, and how often it gets corrected.

  • Salesforce
  • NetSuite
  • Workday
Documents

SOPs, wikis, chats and spreadsheets. Often out of date, but they show how people think the work should run.

  • SharePoint
  • Notion
  • Slack
  • Gmail
Figure 2. Three sources for mapping real work. Systems show what happened; people explain why; documents show how it was meant to work.Source: The Startup Ideas Podcast, with Varick Agents, October 2026

The mix depends on the company. A large enterprise may have ten years of Salesforce history; a smaller one keeps most of its process in people’s heads. System data without interviews misses half the picture, because it shows what happened but not why. And one interview is never enough: one person can tell you the wrong story.

7 → 20

A $5 billion public software company’s process document listed seven steps for getting a quote signed. Process mining on its CRM found 20 steps and seven loops, and 61% of requests looped back to the start.

Source: The Startup Ideas Podcast, October 2026

30%+

The typical gap Varick Agents finds between a team’s written procedure and how the work actually runs. In exception-heavy work, such as accounts-payable exceptions, it is often over 70%.

Source: Daniel Kornum, Varick Agents

That gap is why discovery comes before building. You can’t decide where AI belongs in a process you can’t see.

What skills does an FDE need?

The role is hard to fill because it combines three jobs that rarely live in one person:

The workThe codeThe AIFDE
  • The workHow the business process really runs, and what systems like Salesforce or NetSuite can and can’t do.
  • The codeShipping production software that acts inside those systems, with security, audit logs and governance.
  • The AIChoosing models, writing evals, tuning the harness, and recovering when an agent gets something wrong.
Figure 3. Three skill sets in one person. Plus the ability to explain all of it to executives, in each buyer’s terms.Source: The Startup Ideas Podcast, with Varick Agents, October 2026

On top of that, an FDE has to explain the work in each buyer’s language: days to close the books for a CFO, cycle time and win rates for a CRO, hiring speed for a head of people. On The Startup Ideas Podcast with Greg Isenberg, Varick’s CEO, Vasuman Moza, described someone who is exceptional at all three, and can sell it to senior leaders, as very rare.

Why are FDEs hard to scale?

FDEs work. But every leader running an AI transformation runs into the limits of the model:

  • They are scarce and expensive. Top FDEs are paid like senior engineers or more, and there aren’t many of them.
  • Each engagement covers one team at a time. A company with dozens of departments can’t wait for a few engineers to reach them all.
  • The knowledge leaves with them. When the engagement ends, the map of how work happens starts to go stale.
  • The strategy rests on a few people. If your AI roadmap lives in one engineer’s notes, that is key-person risk.
  • Discovery runs on word of mouth. Interviews capture what people remember and choose to say, at one moment in time.
Figure 4. One team, or every team. An FDE engagement lights up one team for a few weeks. The goal is every team, all the time.

How do you give everyone FDE capabilities?

The FDE playbook doesn’t have to depend on a handful of outside engineers. Its core steps can run continuously, for every team, with your own people at the center:

  1. See the real work from each person’s work on company devices, their email and calendar, and their AI agents’ logs, so the map stays current as the work changes.
  2. Sort every step into delete, plain code, agent or human. See the four buckets.
  3. Prove it on your own work by comparing people, agents, and agents with a reviewer on real tasks. See cost per task.
  4. Share what works as skills, SOPs and agents, so one team’s win becomes everyone’s.

The difference is who holds the knowledge. Instead of outsiders who arrive, automate and leave, your own people become the experts on where AI helps their work, and the know-how stays with the company. That is what it means to make every person forward deployed.

Questions people ask

Is a forward-deployed engineer the same as a solutions engineer?

No. A solutions engineer helps win the deal by showing that a product can work, before the contract is signed. A forward-deployed engineer works after the deal, inside the customer’s environment, and owns making it work in production.

How is an FDE different from an AI consultant?

A consultant mostly advises and hands over a plan. An FDE writes and ships the software, then measures whether it worked. The best FDEs also know what a team needs without making it explain every workflow from scratch.

How long does an FDE engagement take?

It varies. Varick Agents, for example, spends at least four weeks auditing how the work really runs before touching a model, then builds, and goes live over the following months with people still approving the high-stakes steps.

Do we need to hire FDEs to adopt AI?

You need what FDEs do: see the real work, decide where AI belongs, build inside your existing tools, and prove the result. You can hire for it, train your own people in the playbook, or use software that runs discovery continuously across every team.

Why are forward-deployed engineers paid so much?

Because the work sits close to the money. An FDE who removes cost or speeds up a core process can create far more value than their salary, and some private-equity firms give FDEs a share of the value they create.

Sources

  1. “Dev versus Delta: Demystifying engineering roles at Palantir.” Palantir Blog, April 8, 2019.
  2. “A Day in the Life of a Palantir Forward Deployed Software Engineer.” Palantir Blog, November 2, 2020.
  3. Joe Schmidt. “Trading Margin for Moat: Why the Forward Deployed Engineer Is the Hottest Job in Startups.” Andreessen Horowitz, June 4, 2025.
  4. Eyad Khrais. “Forward Deployed Engineering 101.” Varick Agents.
  5. “AI Forward Deployed Engineering: The blueprint to breaking into FDE in 30 days.” Varick Agents.
  6. Daniel Kornum. “If AI is so great, why isn’t it working?” Varick Agents.
  7. Greg Isenberg, with Vasuman Moza of Varick Agents. “Become a $1M/yr FDE (Full Course).” The Startup Ideas Podcast, October 1, 2026.