The student version — published free, in full. This is the short profile written directly to the student; every paid career includes one. Read the full parent profile →
Written for you, not about you.
This one's free — all of it. Send it to whoever you like.
We'll start with the number, because you'll find it anyway.
Recent computer science graduates in the US face an unemployment rate of 6.1%.
Art history graduates: 3.0%. English: 4.9%. Performing arts: 2.7%.
And at the same time, the government projects software developer employment growing 15% between 2024 and 2034 — about five times the average job.
Both of those are true. Working out how is the whole point of this document.
The profession is growing. The way in has contracted sharply.
When an experienced developer says they still get recruiter messages, and a new graduate says they've sent two hundred applications and heard nothing — they're both describing the market accurately. Just not the same market.
Here's what actually happened to entry-level work:
And to be straight with you: not all of that is AI. There's a layoff overhang of 500,000+ experienced engineers, return-to-office rules narrowing where jobs are, and offshoring of junior work. We'd be overstating our own framework if we blamed all of it on the technology. What AI changed is what a junior job contains, not just how many exist.
Every career we scored has six protection factors. Here's software's problem:
==-No licence. There's no professional body, no statute reserving any software activity to a qualified human. Nothing stops a company replacing the work the moment the tools are good enough.==
No physical requirement. It's text in, text out — which is why you can do it from anywhere, and why nothing physical protects it.
Those two sit at zero for everyone in software, at every level, permanently. A principal engineer has the same two zeros as an intern.
So the only protection available is judgment and accountability — and both arrive with seniority. A nurse is protected on day one by law. A developer is protected once someone trusts them with a system.
Which is awkward, because the route to being trusted ran through the work that got automated.
| Track | Score |
|---|---|
| Embedded / safety-critical systems | 4.7 |
| Security engineering | 5.3 |
| Senior engineer / architect | 5.4 |
| ML / AI engineering | 6.0 |
| Backend / infrastructure | 6.8 |
| Frontend / application development | 7.6 |
| Entry-level developer | 8.1 |
Two things to notice. Nothing scores below 4.7 — there's no genuinely safe corner. And the gap between the top and bottom is 3.4 points, which widened faster than almost any profession we measured.
Here's the thing people get wrong, and it matters beyond software.
Being close to AI does not protect you from it.
You'd think understanding these systems would be the safest place to stand. It isn't — and the reason is structural rather than ironic.
==+Exposure is set by the shape of the work, not the subject.== Does it take structured inputs? Produce text or code? Need no physical presence? Carry no licensed accountability? Software answers yes to all four.
A data analyst isn't protected by understanding AI, any more than a copywriter is protected by understanding language.
So the advantage isn't knowing about AI. It's what you do with that.
Debugging genuine novelty. The failure nobody has seen before, with no matching pattern. The single most valuable and least automatable thing in software.
Architectural judgment. Deciding what a system should be — and what it must never do — which requires knowing constraints that aren't in the ticket.
==+Owning something that runs. Being the person who answers when it breaks at 3am.== This is what accountability actually means here, and it's most of the protection.
Verification. Knowing when generated code is subtly wrong in ways that compile and pass tests. That only works on top of fundamentals you actually built yourself — which is exactly why the boring modules matter more now, not less.
If you're good at maths and like building things, this is the most decision-relevant thing in our whole index.
| Software | Engineering | |
|---|---|---|
| Licence? | No | Yes — the PE stamp |
| Need to be physically somewhere? | No | Yes — sites, commissioning |
| Who's liable if it fails badly? | The company | You, personally |
| Entry-level score | 8.1 | 4.0 licensed |
| Graduate market | 6.1% unemployment | ~3 open jobs per qualified candidate |
| Starting salary | ~$87,000 | ~$81,000 |
The salary gap at entry is about $6,000 — smaller than most people assume. Software pulls ahead later, but it pulls ahead for the people who get in, and that's the part that changed.
This isn't an argument against software. Plenty of people should do it and the senior end is well protected. It's an argument that the comparison deserves more than the twenty seconds it usually gets.
Getting the first job is genuinely hard right now. Multi-round technical interviews, a median search running seven to nine months, and by some estimates 200–400 applications per offer.
The tools keep changing. Energising for some people, exhausting for others, and it never stops.
And there's a training problem nobody has solved. Engineers used to build judgment by writing something badly, watching it break, and fixing it. Reviewing generated code might build judgment too — but nobody knows if it builds it as well, and the industry is running that experiment on your entire cohort without a control group.
Which means: seek out the hard problems deliberately. Nobody's going to make you learn the slow way any more.
Not "software" generally. These:
Embedded and safety-critical systems — 4.7. Best score in the field. Regulated, physical, and consequences when it fails.
Security engineering — 5.3. You're up against an opponent actively trying to be unpredictable, which is a completely different problem from complexity.
ML and AI engineering — 6.0. Building the systems rather than using them, with genuinely novel failure modes.
Or software plus a domain — health, energy, finance, defence. The domain is the protection, and those employers see far fewer applicants than the companies everyone applies to.
The honest test: do you enjoy the debugging more than the building?
Because most of the job is finding out why something doesn't work. If that's the frustrating part for you rather than the interesting part, that's worth knowing now.
It also suits people who can be stuck for hours without losing the thread, who are genuinely curious about how things work rather than just what they produce, and who keep learning without being made to.
Harder if you need closure — software is never finished — or if you chose it mainly for the salary. That last group is over-represented among people who leave.
Build something real and ship it. With users, however few.
Not a tutorial project. Something that exists, that someone else uses, that broke and you fixed.
In a field with no licence and no credential, working software is the credential — and almost nobody your age has any.
It also answers a question no amount of thinking will: whether you'll actually do this when nobody's making you.
Ship something. See above. It's the highest-leverage thing available to you right now.
Then find someone two or three years into a software job and ask what they actually did in year one — and whether that job still exists in the same form. Their answer will be more current than anything written down, including this.
For twenty years, "learn to code" was the advice you could give any teenager without knowing anything about them. That's what stopped being true — not because the field is dying, but because the gap between its best and worst outcomes got very wide, very fast.
So computer science isn't a bad choice. It's a bad default.
And here's the part that's still true: software is one of very few careers where you can build a working thing, alone, with nothing but a laptop and time — and nobody's permission is required to try.
That's rare. It's also exactly how you get to the protected end.
If you want to go further — three things worth twenty minutes
The Labor Market for Recent College Graduates — the Federal Reserve tracks unemployment and pay by major. Look CS up yourself, then compare it with engineering and nursing.
Computer science grads are struggling to find work — where the 6.1%-versus-art-history comparison comes from.
Software Developers — Occupational Outlook Handbook — the government's own 15% growth projection. Read it against everything above; the tension between them is the real story.
The full computer science profile is free too — all sixteen sections, the scoring in detail, what to ask a program, and where we think we might be wrong. We publish one in full so you can see what the paid ones look like before anyone spends anything.