The Return of the House Call: What 'Forward Deployed' Really Says About AI
Movement 1: The Raw Feedback of the House Call
More than a decade ago, working for Network Solutions meant my teams and I would visit customer workplaces. We weren't there to install a new feature; we were there to see the real pain—the kind of network friction and workflow bottlenecks that never showed up in a clean report. We were listening for the ouches and the uh-ohs.
This idea—embedding deeply to understand the customer's mess—is not new. It's a core component of building products that truly resonate. The tension then, and now, was always the same: Do you spend your time solving that customer's immediate problem, or do you use the visit to solve the collective problem for everyone?
Movement 2: Two Competing Pressures
When a company today says "we're hiring forward-deployed engineers," they are managing two fundamentally different motivations:
- Roadmap Oxygen (The Strategic Motive): This is the long game. It's admitting that the only way to get reliable, high-quality signal for the product roadmap is to have brilliant engineers sit inside the customer's messy reality. Palantir perfected this model—at one point having more forward-deployed engineers than traditional software engineers—using field insights to continuously evolve their platform.
- Time-to-Value Adrenaline (The Survival Motive): This is the financial pressure. The current spike in demand for these roles signals that this motive is winning. As Andreessen Horowitz observed in their analysis of AI startups, companies are "trading margin for moat"—sacrificing short-term profitability to prove their AI products can create immediate value. When OpenAI established its Forward Deployed Engineering team in early 2024, they weren't just thinking about product feedback; they were responding to enterprise customers who needed their complex AI systems integrated quickly.
When technology races ahead of users' ability to wrangle it, the need for these on-site "translators" becomes critical, turning a product problem into a service necessity.
Movement 3: The Uncomfortable Question About Product Maturity
This leads to a sharper question: If a company requires armies of forward-deployed engineers for years just for implementation, what does that quietly signal about the underlying product?
A truly scalable product achieves "self-serve and bulletproof" status. It handles the long tail of real-world use cases without constant customization. If you need a perpetual stream of high-paid, code-savvy problem-solvers, you may be admitting that your core offering still lacks product-market fit for most customers.
Consider Palantir's evolution. Until 2016, they had more forward-deployed engineers than traditional software engineers. Then they launched Foundry—a more productized platform—and many FDEs transitioned back to core engineering, bringing their field experience to build better abstractions. This suggests that massive FDE teams can be a transitional phase, not a permanent state.
But here's the economic reality that complicates this narrative: The margins on AI consulting can be better than pure software margins. A former Palantir engineer candidly observed that true forward deployment requires accepting chaos, overlapping work, and "burnt-out husks of FDEs" as the cost of learning. Companies pursuing this model need high prices—enterprise or government customers who can afford $200K+ per engineer annually.
The Palantir Exception and the AI Paradox
Palantir has been running this playbook for over 20 years. Is that a feature or a bug?
Product leader Marty Cagan argues that Palantir's genius isn't just sending FDEs—it's their platform strategy. Each forward-deployed engineer builds prototypes using the latest platform capabilities, while the core product team constantly works to generalize and incorporate new capabilities identified in the field. The platform evolves through structured feedback loops.
But most companies claiming to do "forward deployed engineering" lack this discipline. They're using the label as a palatable way to package what used to be called: "We're still in beta, but we'll customize it for you if you pay enterprise prices."
The AI wave has intensified this pattern. As one AI startup engineer put it, being a forward-deployed engineer means being "an engineer, salesman, customer support representative, and model performance engineer all in one person." OpenAI, Anthropic, Scale AI, and dozens of AI startups are all hiring these roles—not primarily for roadmap feedback, but because LLMs are genuinely difficult to integrate into complex enterprise workflows.
The Question That Should Keep Founders Up at Night
What would need to change—either in the technology or in company incentives—for the primary reason to swing back toward "deep feedback for roadmap" rather than "survival through immediate value delivery"?
Every founder and engineer stepping into this cycle should ask themselves:
"If your product tomorrow became magically self-serve and bulletproof, would you cheerfully disband your forward-deployed team... or would you miss the richest feedback firehose you ever had?"
The honest answer reveals whether you're building a services business disguised as a product company, or a product company that's genuinely learning from the field.
The forward-deployed model isn't inherently wrong. But if after three years you're still hiring more FDEs than product engineers, you're not building toward scale—you're admitting that your product's complexity is a feature, not a bug. And in a world where investors expect AI products to achieve software economics, that's a dangerous admission to make.
Sources: This essay draws on reporting from Pragmatic Engineer's analysis of forward-deployed engineering roles, Palantir's public job descriptions and engineering blog, OpenAI's career postings, Andreessen Horowitz's "Trading Margin for Moat" analysis, Marty Cagan's SVPG analysis of FDE culture, and firsthand accounts from engineers at Palantir, Baseten, and other AI companies.
Originally published at shashi.co
Disclaimer: This blog post reflects my personal views only. AI tools may have been used for brevity, structure, or research support. Please independently verify any information before relying on it. This content does not represent the views of my employer, Infotech.com.
Comments