AI and software engineering jobs: an honest read

The honest read is this: AI is changing how engineering work gets done faster than it is changing how many engineers companies want to hire. The headlines that say one of those things will erase the other are reading a real shift through the wrong lens. What is actually moving is the mix. Companies still raise rounds and still need to build, but the kind of engineer they reach for first has shifted toward people who can take ambiguous problems from zero to shipped, with or without a model writing some of the code.

Most of what gets written about this is either doom or hype. Both are easy to write and not very useful when you are deciding where to interview next quarter. This post sticks to what we can see in the roles.cc board and in how funded startups actually staff up after a raise, and it is candid about the parts nobody can measure yet.

What is actually changing in demand?

Two things are clearly moving, and they pull in opposite directions, which is why the net picture looks confusing.

The first: AI coding tools have made the median engineer measurably faster at the well-defined middle of the job. Boilerplate, test scaffolding, a CRUD endpoint, a migration, a first-draft refactor. Work that used to take an afternoon often takes an hour. When a unit of output gets cheaper, the companies that were on the fence about a feature now build it, which creates more work, not less. But the same shift means a team of 6 can now cover what a team of 9 covered in 2022. Whether headcount goes up or down depends entirely on whether the company has more it wants to build than it can fund. Funded startups almost always do.

The second: the floor moved. The simplest version of an engineering task (translate a clear spec into clean code) is the part a model does best. So the roles that were mostly that are the ones under pressure. The roles that were mostly judgment (what to build, why, what breaks at scale, what the data actually says) are not. This is why early-career hiring tightened first and senior hiring stayed strong. We wrote about the timing of that in the best time to job search as a software engineer.

What is not changing?

More than the headlines admit. A model can write a function. It cannot decide which function is worth writing, sit in a sprint planning argument, own an outage at 2am, or carry the context of why the last three decisions were made. Startups do not pay senior engineers for typing speed. They pay for the judgment that prevents six weeks of building the wrong thing.

• Ownership. Someone has to be accountable when the system is down. That has not moved an inch.
• Judgment under ambiguity. Deciding what to build, and what not to, is most of the job at a startup. Models do not have the company context to make that call.
• Systems thinking. Knowing why a clean-looking change will melt the database at 10x traffic is still scarce and still expensive.
• Trust and review. More generated code means more code that needs a human who can tell good from plausible. That makes review skill more valuable, not less.
• Founding-engineer work. Taking a vague idea to a shipped product touches all of the above. We broke this down in what a founding engineer actually does.

Startups do not pay senior engineers for typing speed. They pay for the judgment that prevents six weeks of building the wrong thing.

Where is hiring concentrating?

Demand is not evenly spread, and the gradient is steeper than it was. Here is where the funded roles are actually clustering, based on what comes through the board.

Notice the pattern: the squeeze is at the narrow, well-specified end, and the pull is at the broad, judgment-heavy end and the AI-building end. The funding signal sharpens this. A company that just closed a round has a specific, approved hiring plan, and right now that plan disproportionately includes someone to own an AI feature or someone senior enough to make a small team go fast. You can watch which companies are in that window on the recent raises page. For the mechanics of why a fresh raise is the cleanest hiring signal, see why funding recency is the best hiring signal.

Does this mean fewer engineering jobs overall?

Not at funded startups, and that is the slice that matters if you are reading this board. A startup that raises a Series A is buying about 18 to 24 months of runway and a mandate to grow. Cheaper per-feature output does not make a growing company want to build less. It makes it want to build more with the budget it has, which usually means hiring strong people and giving them larger surface area, not freezing the team.

The honest caveat: this is clearest at the startup end and murkier at large companies, where AI savings can show up as a hiring slowdown rather than more shipping. We place engineers at venture-backed startups, so that is the picture we can speak to with confidence. If you are weighing the two worlds, startup vs big tech for a software engineer walks through the tradeoffs.

How do you stay valuable through this?

The answer is not to out-type a model. It is to be the person the model makes more productive instead of the person it makes redundant. Concretely:

1. 01Move up the stack of judgment. Get good at deciding what to build and why. The closer your work sits to product and customer impact, the harder it is to automate.
2. 02Get fluent with the tools, not afraid of them. An engineer who ships a feature in two days using AI tooling well beats one who takes a week refusing to. Treat the tools as a force multiplier on your judgment, not a threat to your job.
3. 03Own something end to end. Take a domain from spec to production to on-call. End-to-end ownership is the single most automation-resistant shape of work.
4. 04Sharpen review and systems sense. More generated code means the rare skill is telling correct from plausible, and seeing what breaks at scale. Lean into it.
5. 05Build in the AI layer if you can. The fastest-growing demand is for people who can build and serve models in production. Even adjacent experience reads well right now.

None of this is new advice dressed up. It is the same thing that made engineers valuable before, with the dial turned up. The part of the job that was always the real value (deciding, owning, judging) is now a larger share of the value, because the routine part got cheaper. If you want a read on how to present that value in a search, how to evaluate a startup job offer covers what to weigh once the conversations start.

A worked example

Say a startup raises a $12,000,000 Series A and plans to grow engineering from 5 to 9 over the next 18 months. In 2022 that plan might have been 5 to 12 for the same roadmap. The team is smaller for the same scope because tooling stretched each person further. But it is still 4 net new hires, and the company will reach for senior generalists and one ML-leaning engineer to own its AI feature, not 4 narrow junior reqs (illustrative, not advice). The total number went down. The value of the specific people they want went up. That is the shift in one example.

For the candidate, the takeaway is direct: the bar for the roles that exist is higher, and the comp for clearing it is strong. Senior pay at funded SF and NYC startups has not softened with this shift. We keep current numbers in senior software engineer salary in SF and NYC for 2026.