You open a product manager listing and it asks for Python. You open an engineering listing and it asks for customer discovery. Someone you used to work with has a new title, “AI deployment lead,” and nobody can explain how it differs from the old one.
The scramble feels random. It is not. Job titles in tech have been merging for fifteen years, and every merge followed the same logic. That logic also explains why the forward deployed engineer (FDE), a role Palantir built long before ChatGPT, has become one of the fastest-growing jobs in AI.
Underneath it sit two questions we have tackled before: whether AI will replace jobs at all, and why harness engineering is replacing prompt engineering. Both are really questions about the same pattern, and that pattern ends at the FDE.
Key takeaways
- Roles tend to merge when the handoff between two specialists becomes the most expensive part of the work.
- DevOps, the data scientist and the product builder each removed one handoff. Prompt engineer is the counterexample: a title that became a skill.
- AI makes building cheap. The bottleneck is now making software work inside a customer’s organization, and that is the FDE’s job.
- The merging is uneven, so chase the skill bundle, not the title.
Four turning points on the way to the FDE
For decades, software ran like a relay race. Analysts gathered requirements, developers wrote code, testers checked it and operations ran it. Each handoff cost time, which was a fair trade while every step was slow and hard. The four turning points below show what happens when the steps get fast and the handoffs become the problem, and what happens when a title bets on a single tool.
2008: the data scientist
Companies had more data than anyone could use, and the work was split across statisticians, engineers and analysts. According to a 2012 Harvard Business Review article by Thomas Davenport and D.J. Patil, Patil and Jeff Hammerbacher coined the title “data scientist” in 2008 while leading data efforts at LinkedIn and Facebook. Historians dispute whether it was the first use, since earlier uses exist. The point is what the title bundled: statistics, software engineering and business communication in one person, because passing a question across three specialists was too slow.
2009: developers and operations
In June 2009, John Allspaw and Paul Hammond gave a talk titled “10+ Deploys a Day: Dev and Ops Cooperation at Flickr.” A consultant in Belgium named Patrick Debois watched the stream at home, and it clicked. He organized his own conference that autumn, and DevOps had a name.
The idea was simple. The wall between the people who wrote code and the people who ran it was the real problem. Developers who owned what happened after they shipped made better software.
2023: the job that became a skill
Then generative AI arrived, and “prompt engineer” was heralded as tech’s hottest new role. Fast Company later reported that it was fading as a standalone title, with an Indeed economist saying it was typically being folded into roles like machine learning engineer or automation architect. The skill stayed. The job did not. If you want a snapshot of the peak, we published a 10-step guide to becoming a prompt engineer in 2023.
That is the warning in this story: narrow, tool-shaped titles can dissolve into the roles around them.
Now: the product builder
The latest merge is inside product teams. When one person can draft a spec, prototype a design and generate a working first version in an afternoon, the classic split between product manager, designer and engineer gets expensive.
LinkedIn is the clearest public example. It replaced its associate product manager program with an associate product builder program that teaches coding, design and product management together. Its chief product officer, Tomer Cohen, told Lenny’s Podcast the goal is people “who can flex across” functions, according to Business Insider.
Analysts see the same direction. Gartner expects software teams to become smaller, more autonomous and more product-focused, with role boundaries beginning to disappear. Bain describes a move from the classic squad of one product manager and six to eight engineers toward smaller hybrid-agentic pods where engineers work alongside agents.
The pattern: when the handoff gets expensive, roles merge
So why does this keep happening? Line up the merges and the logic is the same each time: a role fuses with its neighbor when the boundary between them becomes the most expensive part of the work. Prompt engineer is the exception that proves the rule. It never removed a handoff, so nothing held the title together. The table is my analysis, not a formal model.
| Merge | What combined | The handoff that got too expensive |
|---|---|---|
| Data scientist | Statistician, engineer and analyst | Getting from raw data to a business answer |
| DevOps | Developer and operations | Getting written code running in production |
| Product builder | Product manager, designer and engineer | Getting from an idea to a working prototype |
| Forward deployed engineer | Builder and customer | Getting working software to work inside a real organization |
The last row is the one AI has made urgent.
Why the FDE is the next merge
So why the FDE, and why now? When AI cuts the cost of building, the cost of everything around building stands out. A model can write the code. It cannot sit inside a bank and discover that the data lives in three systems that disagree about what a “customer” is, that an undocumented permission blocks the whole workflow, or that the people who will use the agent do not trust it yet. Call it the deployment gap: the distance between software that works in a demo and software that works in an organization.
The FDE closes that gap. They scope the use case, wire an AI agent into the customer’s systems, test it against real workflows and iterate until it sticks. Think of an engineer, a consultant and a product manager in one person, with the customer’s problem as the roadmap.
None of this was invented for AI. Palantir built the model early and, according to several accounts, employed more forward deployed engineers than traditional software engineers well into the mid-2010s.
What changed is who is copying it. The Financial Times reports that OpenAI, Anthropic and Cohere are all expanding their FDE teams. Fortune, citing Lightcast data, reports that postings for forward-deployed roles rose more than 1,000% between January and August 2026 versus the same period a year earlier. Other trackers report smaller jumps, so trust the direction more than the exact percentage.
And notice what the job actually is. It is not generating code. It is deciding how much freedom an agent gets, what happens when it is wrong, and who answers for it. That is the same judgment behind graduated autonomy, where an agent’s permissions are tiered by action risk.
Is it hype? Three caveats
- The title is inconsistent. KDnuggets asked whether the FDE is AI’s hottest new career or consulting with a better title. Some commentators think startups are copying Palantir without its underlying model, and Palantir’s own leadership has said many imitators are struggling.
- Merging is uneven. A five-person startup has always asked people to do several jobs. A bank or hospital system has compliance and scale constraints that keep specialists in place, and Bain cautions that no single structure suits every organization. LinkedIn is one company, not a census, and Gartner’s tiny-team model is a forecast.
- Specialists do not disappear. Deep expertise in security, data infrastructure, research or design still matters. What changes is that depth now needs a layer of breadth around it.
What this means for your career
Where does that leave your career? Titles are snapshots of where the handoffs happened to sit at one moment. Data scientist and DevOps engineer each looked permanent when they arrived, and prompt engineer looked permanent for about as long as a news cycle.
That makes the product manager question easier. A VentureBeat essay argues the old ratio of one PM to eight engineers now behaves more like one to twenty, because engineers with coding agents ship faster than discovery work can keep up. That is one writer’s argument, not settled data, but the direction is plausible. If you want a PM title for the seniority, that is a weak bet. If you want to do the PM job, deciding what to build and why, that skill is worth more than ever. It just no longer belongs to one box on an org chart.
Four capabilities do the most work when one person directs several agents and owns an outcome:
- Judgment. Agents produce plenty of plausible output. The scarce skill is knowing what is right and what to throw away.
- Systems thinking. Understand how data, permissions, workflows and people connect. Knowing which layer of an agent is failing, the reasoning loop or the harness around it, is a good example of the habit.
- Domain depth. Be genuinely strong in one area, such as engineering, data, design, finance or healthcare. Breadth without depth is what agents replicate cheaply.
- Comfort shipping. Put a working version in front of real users quickly and iterate. A record of shipped outcomes travels across job titles better than any job description.
Go back to those job listings. A product role that asks for Python and an engineering role that asks for customer discovery are the same signal: the handoffs are collapsing, and teams want people who can carry a problem across them. The titles will keep changing. That skill is the part that lasts.
Where to start
Judgment and shipping are hard to build from reading alone. The FDE role is learned by doing it, under real deadlines, against real client problems. That is the idea behind our forward deployed engineer bootcamp: a 12-week, full-time residency in Redmond/Seattle, WA where you build and deploy AI systems for client-style problems, then interview with hiring partners. It is built for recent engineering graduates and experienced IT professionals, and it is free to cohort members because hiring partners fund every seat. Admission is competitive, with 30 seats per cohort.
The next cohort starts January 18, 2027, and applications close December 21, 2026. See the curriculum and apply.
Want to build AI agents that can reason, plan, and execute autonomously?
Learn more