The Translation Layer Is Gone: What Ajey Gore Gets Right About Roles After AI

Ajey Gore's "AI ate my role! What's next?" reads like a sequel that knows exactly which questions its predecessor left unanswered. After "The Anatomy of an AI-Native Org" described what happens to the shape of companies when agents absorb the work that used to flow between humans, the obvious follow-up was personal rather than structural. Not what happens to the org chart, but what happens to me. The piece answers that by walking through seven roles, but the walkthrough is almost beside the point. The real argument is a single distinction, and once you see it, every role section becomes a corollary.

The thesis underneath all seven roles

Gore's central claim is that every job in software splits along the same fault line. On one side sits translation: converting a well-defined input into a well-defined output, running a process that yields a predictable artifact, coordinating other people's work. On the other side sits judgement: deciding what correct means when the situation is ambiguous, holding context across a problem larger than any single ticket, taking responsibility for an outcome rather than an output.

The translation work collapses into the agent layer. The judgement work grows. This is the entire argument, and its power comes from being indifferent to job title. A Product Manager and a junior engineer and a QA lead are all running the same internal audit: which parts of what I do are translation, and which parts are judgement? Gore's prescription for the next five years is whatever moves you decisively into the second column.

What makes this more than a motivational slogan is the honesty about cost. Gore distinguishes between roles that evolve, where the work sharpens but stays recognizable, and roles that transform, where the job becomes a different job wearing the old name. That second category is where the discomfort lives, and he refuses to paper over it.

The roles that sharpen

For Product Managers, the argument exposes a fiction the industry maintained for a decade. Product and project management shared a title in most companies, Gore argues, but never shared a capability. He estimates the traditional PM spent roughly 70% of the week on translation work, PRDs, research synthesis, prioritization scoring, alignment decks, and that this 70% was project management wearing product clothing. Agents now do it faster and at higher fidelity than any human stitching a status update together on a Friday afternoon. When that 70% vanishes, the gap it leaves is precisely the gap between coordinating a process and having a thesis.

Gore offers four surviving lanes, vision, intuition, research, and customer proximity, but the deeper move sits above all four. The new PM becomes a business owner held explicitly accountable for a number, revenue or margin or retention, rather than for features shipped. The conversation with leadership stops being "did you ship" and becomes "did you move the number." His observation that intuition is what tells you when the agent's output is plausibly wrong is the sharpest line in the section, because it reframes taste not as decoration but as an error-detection function that becomes more valuable precisely as machine output becomes cheaper.

Senior and staff engineers fare best in Gore's reading, because their work was always nearer to judgement than translation. What disappears is the boilerplate, the rote refactor, the hand-tooled scaffolding. What grows is leverage. He retires the mythical 10x engineer and replaces it with the 100x pattern, one senior engineer directing agents to do what once took a team of eight. The catch is psychological rather than technical. Writing the code was the comfortable part, the part with visible output and immediate dopamine. The new work, specs and evals and architectural review of agent-generated code, produces fewer trophies and more leverage. Gore is clear that this trade is worth making and equally clear about why people resist it.

Project and Program Managers face the most aggressive version of the evolution story, aggressive enough that it arguably belongs in the transformation column. Their work was almost entirely translation: status synthesis, sprint facilitation, roadmap-to-JIRA conversion, standup chairing. Agents reading the PR graph and deploy log produce more accurate status than a human can assemble by hand. Gore names three surviving lanes, each of which asks the PM to become something adjacent: more product (taste), more quality engineering (harness), or more governance (boundary work on compliance and legal). The blunt addendum he insists on saying aloud is that the total number of seats will shrink, and the survivors will be senior, lane-specialized, and already six months into the shift.

Engineering Managers get the simplest diagnosis in the post. The EM who has not coded in five years is the most at-risk profile; the EM who still ships something every quarter is the most resilient. Standup facilitation and status writeups collapse. Design contribution, hiring, harness ownership, and incident command grow. The instruction is to get hands back on the work, enough to be in the design conversation rather than merely chairing it.

The roles that become something else

The transformation sections are where Gore's writing earns trust, because he stops selling and starts admitting what nobody has solved. The junior engineer problem is the one he calls the hardest career conversation in software, and he does not pretend otherwise. The traditional path to senior engineer ran through the translation work itself: converting tickets to PRs, shipping bugs, absorbing feedback, slowly developing judgement over five or seven years. The agent now does that conversion work faster than any junior can, which means the apprenticeship pipeline that manufactured senior engineers is breaking, and no replacement exists yet. "Anyone who claims they have one is selling something" is the kind of sentence most thought-leadership posts cannot afford to write.

His advice to juniors is concrete in a way the rest of the post sometimes is not. Do not compete with the agent on execution speed; compete on judgement. He suggests reading three pull requests a week and asking the same three questions of each, what constraint was this solving, what did it trade off, would I have made the same call, and writing the answers down until you have a notebook of critiques that almost no other junior possesses. The reading-and-judging muscle is the senior-engineer muscle, and Gore's claim is that agents have made it more accessible, not less, because there is suddenly far more code to read and judge.

Designers, he argues, are losing production volume, the wireframes from a brief, the mockup variants, the design-system implementation, while keeping taste, brand judgement, system coherence, and the opinionated calls about what good feels like for a specific product. His instruction to stop optimizing a portfolio for output and start building "judgement artifacts," rationale documents, critiques, brand voice guides, is a genuine reframing of what a designer's evidence of competence looks like.

The QA section is the one where Gore is most willing to puncture a comforting narrative. The easy story says QA was undervalued and now gets its moment. He rejects that. For thirty years, he argues, QA created a quiet co-dependency with engineering: developers stopped worrying about quality at the moment of writing because a safety net waited downstream, and quality became something that happened to the work rather than something that lived inside it. When eval suites and automation become load-bearing, the three-week manual QA pass on a feature an agent shipped in three hours simply stops being geometrically possible. The role does not get elevated; it gets relocated, into engineering rather than alongside it, designing evals instead of running them, owning acceptance criteria instead of validating against them after the fact. The differentiator becomes deep domain knowledge about how a specific system breaks under stress.

Customer Success, in Gore's telling, quietly becomes one of the most strategically important seats in an AI-native company. Tier-one support, triage, and FAQ responses go to agents. What survives is product feedback synthesis, the work of distilling hundreds of customer conversations into a point of view about what to build next. The CS people who make the jump stop counting tickets closed and start writing product briefs from customer patterns, showing up to the product meeting with an argument rather than a summary.

What the framework illuminates, and what it leaves open

The strength of Gore's piece is its refusal to let the translation/judgement cut stay abstract. By forcing every role through the same filter, he produces a diagnosis that feels less like prediction and more like a description of pressure already arriving. The honesty about the junior pipeline, the bluntness about shrinking PM headcount, and the rejection of the QA-victory-lap story all signal someone more interested in being right than in being reassuring.

The framework does invite harder questions than it answers. If the apprenticeship layer that produced senior engineers is genuinely breaking, then the supply of people capable of the judgement work, the work everyone is being told to flee toward, is itself under threat. The 100x engineer plus a fleet of agents is a compelling image precisely because that engineer accumulated judgement through the old pipeline that is now closing behind them. Gore acknowledges this directly, saying the industry risks losing a generation and that he is working on the problem, but the tension between "move toward judgement" and "the thing that builds judgement is disappearing" is the unresolved knot at the center of the whole argument. It is to his credit that he names it rather than smoothing it over.

There is also a quieter assumption worth examining: that judgement is durable in a way translation is not. That holds for now, when models excel at well-specified conversion and struggle with ambiguous ownership. Whether the boundary stays where Gore draws it, or whether the next few years see judgement work itself begin to fragment into specifiable pieces, is the question his framework cannot settle from inside 2026. The cut he describes is real today. Treating it as permanent is the one move in the essay that asks for more faith than evidence.

What Gore captures better than most writing on this subject is the texture of the transition as a lived thing rather than a market trend. The defensiveness of the EM who asked, the quiet fear of the junior asking a question they dread the answer to, the PM who shone at coordination and now has to decide whether they ever actually had a product instinct. These are not abstractions. They are the specific moments where a person realizes the comfortable, visible, dopamine-rich part of their job was the part the machine wanted, and the part that remains is the harder half they had been quietly avoiding. His closing line, that AI did not eat the judgement work and cannot, is less a comfort than a challenge: the part of your role that survives is the part that was always the actual job, and now there is nowhere left to hide from it.