For years I introduced myself as an experience director. Lately I say forward-deployed engineer. The honest version is that the title changed but the instinct never did. Maybe I always had the FDE in me: the part that wants to sit inside the messy reality of a client's problem and build my way out of it, rather than hand over a deck and hope someone else feels what I felt.
What has actually changed is the cost of building. When writing software was expensive, the scarce thing was execution, so that is what we organized around. Now the model writes the code, drafts the copy, fills in the boilerplate. The constraint has moved. The bottleneck is no longer "can we build it" but "should we, and how should it feel."
Cat Wu, who leads product for Claude Code at Anthropic, named this shift cleanly. As code gets cheaper, she points out, the valuable skill becomes deciding what to build at all, and then the harder question underneath it: "How can we make users feel most delighted?"
That question does not automate. It is not a ticket you can close. It is the kind of judgment experience people have always carried: knowing something is off before you can articulate why, and trusting that signal enough to act on it. That instinct is built slowly, from years of watching how real people respond to real products, and it is hard to transfer precisely because so much of it lives below language. A model can imitate the patterns it has seen, but it has no stake in whether the result actually moves someone. Someone with that stake still has to make the call.
Here is what I keep seeing in real engagements. Automation eats the legible eighty percent quickly and convincingly. The demo looks great. Then you hit the last stretch, the part that resists specification, where the work stops being about whether something functions and starts being about whether it lands. A flow that technically works but feels cold. A response that is correct but tone-deaf. Output that is ninety-five percent right, which in practice means a human still has to watch the failing five percent, which means it was never really autonomous at all. That last stretch is exactly where experience lives, and it is precisely the part you cannot delegate if you do not have taste yourself.
Cat made a related point about working with agents that has stuck with me: "It is extremely hard to manage agents if you can't do the job yourself." You cannot curate what you cannot recognize. You cannot distill an experience down to its delightful core if you have never felt the difference between fine and right. The model is a great collaborator, but it is reaching for an average of everything it has seen. Someone has to stand at the end of the pipeline and decide that the average is not good enough.
This is the quiet revenge of the experience director. For a while it looked like our craft was being automated. What is being automated is the production. The judgment is not. If anything, it is more concentrated, more leveraged, more valuable, because one person with real taste can now shape a lot of output. Feeling, judgment, the willingness to say "no, that is not delightful yet" and know why: this is what drives the best products, and it always has.
Being deployed forward, in the end, is not really about writing the code. It is about being close enough to the problem to feel what is wrong, and clear enough about what good feels like to curate the machine toward it. The tools got faster. The taste still has to come from a human standing in the room.
Inspired by Cat Wu (Head of Product, Claude Code and Cowork at Anthropic) on Lenny's Podcast: "How Anthropic's product team moves faster than anyone else." Listen here: open.spotify.com/episode/7wTqD5zwl8ZitUeplapvBG
Esben Sohl is a Partner at Gneis, working across brand strategy, product design, and AI. Connect on LinkedIn.
