Palantir coined the role FDE fifteen years ago for forward deployed engineer, basically an engineer embedded inside customer environments, building production systems shoulder-2-shoulder with the people who use them. this year, the role exploded. OpenAI raised $4B to launch an entire deployment company¹. Anthropic partnered with Blackstone for the same play². Google is hiring hundreds of FDEs, job postings grew 800% in under a year.
the thesis is simple: AI made code cheap, but context is still expensive. while anyone can generate software now, the hard part is making it work inside a real company, with real data, real constraints, real humans.. who do things differently than you expected. that's what FDEs do. They sit next to the problem.
wild guess about what nobody is saying yet: the same thing is about to happen to design.
--
code got commoditized. design is next.
41% of code pushed to production globally is now AI generated³. codex, cursor, claude code turned writing software from a craft into an editing job. The engineer's value moved upstream, from syntax to judgment.
i feel like design is following the exact same curve. with figma make, google stitch, v0... describe what you want, get a high fidelity UI in seconds.
but producing a screen is becoming trivial. producing the right screen for the right person in the right context, that's still hard.
--
the tailored enterprise
something fundamental is shifting in how companies buy software, classic model: you pick a saas product, you bend your workflows to fit it, you pay per seat forever.
ever heard about a CRM being so generic that it requires an army of consultants to make it work for your specific business? you pay for the software, then you pay again to make it actually yours.
but when building software becomes cheap, that math breaks. why pay $300/seat/month for something that's 70% what you need when you can build 100% of what you need for less?
Twenty⁶ is an early signal of this shift. as oss CRM designed to be shaped by each team that uses it. a CRM building toolkit. And it's what enterprise software starts to look like when production costs collapse toward zero.
--
where the designer sits
forward deployed engineer exists because code needs context. someone has to understand the customer's data pipelines, their compliance rules, their legacy system that runs everything, the way their team actually works versus how they say they work.
i believe design needs the exact same thing.
every company I've worked with has different users, different mental models, different tolerances for complexity. a fintech dashboard and a logistics dashboard might share the same technical stack but require fundamentally different design decisions. spacing, density, hierarchy, language, the way errors are surfaced... all of is shaped by who is sitting in front of the screen and what they're trying to do under pressure.
AI can generate a dashboard in seconds. tt cannot know that your ops team processes 400 orders before 9am and needs the critical exceptions to scream louder than everything else. It cannot know that your sales team ignores any metric that isn't on the first screen they see. It cannot know that your compliance officer needs an audit trail visible at all times or the tool is dead on arrival.
That knowledge is the designer's job. And it lives on-site, in the room, next to the problem.
--
generative UI changes the game
CopilotKit's generative UI work⁵ is worth paying attention to. They've defined three levels of how AI can control interfaces:
The first is controlled: the developer builds all the components, the agent just picks which one to show. The second is declarative: the agent returns a structured spec (a card, a list, a form), and the frontend renders it with its own styling. The third is open-ended: the agent returns an entire UI surface. Each level trades more control for more flexibility.
This matters because it means the interface is no longer static. It can adapt per user, per context, per task. A support agent sees a different view than a manager. A first-time user sees a different onboarding than a power user. Not because someone designed twelve variants in Figma, but because the system generates the right interface in real time.
But someone still has to define what "right" means. Someone has to build the component library the agent selects from. Someone has to write the design system the agent respects. Someone has to set the constraints so the generated UI doesn't feel like a slot machine.
Google seems to agree. they open-sourced DESIGN.md⁷, a specification for describing your visual identity to AI agents. One file, machine-readable design tokens + human-readable rationale. You write it once, every AI tool reads it. The agent generates on-brand output automatically.
The designer doesn't draw every screen anymore. The designer defines the system, the rules, the taste, then lets the machine execute within those boundaries. That's a fundamentally different job. And it's a more valuable one.
--
the forward deployed designer
So here's where it lands.
Enterprise software is moving from one-size-fits-all SaaS toward tailored systems built per company. The tools exist: open-source foundations like Twenty, generative UI frameworks like CopilotKit, AI builders like v0 and Lovable, design specifications like DESIGN.md. The production cost is collapsing.
What's not collapsing is the cost of understanding a customer well enough to build the right thing. The cost of sitting with a team, watching them work, noticing what they ignore and what they reach for, translating that into software that fits like it was made for them, because it was.
The forward deployed designer is the person who does this. Not a consultant who drops a Figma file and leaves. Not a UI designer who works from a requirements doc three layers removed from the user. Someone embedded in the customer's world, with the tools to build and ship directly, who understands that design is not about making things look good. it's about making things work for specific people in specific contexts.
The FDE boom proved that code without context is worthless. The same is true for design. Maybe more so.
If code is the commodity, and AI handles the production, what's left is the hard part: knowing what to build and for whom.
That's design. Always has been.
- OpenAI launches the Deployment Company, $4B+ subsidiary backed by TPG consortium
- Anthropic partners with Blackstone, Hellman & Friedman, and Goldman Sachs, ~$1.5B enterprise AI services firm
- 41% stat, AI in Design 2026
- Google Stitch, AI-native UI design canvas
- CopilotKit generative UI, three patterns for AI-controlled interfaces
- Twenty, open-source CRM
- DESIGN.md, open-source design specification for AI agents