Pivotum Profile/Computer science/Fall 2026

The complete profile — published free, in full, so you can judge the depth before buying anything. There’s also a free student version →

Is a Computer Science Degree Still Worth It?

The Complete Computer Science Profile

Pivotum Degree Risk Index — Fall 2026 Edition

This profile is free. We publish one in full so you can judge the depth before buying anything. It is the same length, structure and standard as every paid profile in the index.


Before you read this — a note for parents

You're likely reading this before your child is. That's normal, and it's worth pausing on how you use it.

Don't lead with the scores. If you open the conversation with a risk number, one of two things happens: they get frightened and shut down, or they get defensive about a decision they've already made. Neither produces a better choice.

The better approach: read this yourself first, then share it with them as information you both get to react to, not as a verdict you've reached.

What this is. A facts document, not a recommendation. We've set out what's known, where the evidence is contested, and where we might be wrong.

Three things to keep in mind:

  1. They're choosing, not you.
  2. This is a starting point, not a ruling.
  3. Don't do it all at once.

A practical note: there's a separate short version written directly to them. Send them that one.

And one thing specific to computer science. This is the profile most likely to cause an argument, because the advice your child has heard their entire life — learn to code, do something with computers — is the advice this document questions. Handle that carefully. A teenager who has built an identity around being the technical one does not need to be told they were wrong. They need to understand that the field split, and which part of it to aim for.


The short answer

Computer science is not dying. Its entrance is.

Recent computer science graduates in the United States face an unemployment rate of 6.1% — higher than art history at 3.0%, English at 4.9%, and performing arts at 2.7%. Meanwhile the Bureau of Labor Statistics projects software developer employment to grow 15% between 2024 and 2034, roughly five times the average across all occupations.

Both of those numbers are real, and the contradiction between them is the entire story.

An entry-level developer scores 8.1 out of 10 for AI exposure. A senior engineer or architect scores 5.4. Embedded and safety-critical work scores 4.7.


The scores

Track20232025Fall 20263-yr movementBand
Embedded / safety-critical systems4.04.34.7+0.7Moderate
Security engineering4.65.05.3+0.7Moderate
Senior engineer / architect4.65.05.4+0.8Moderate
ML / AI engineering5.05.56.0+1.0Moderate
Backend / infrastructure5.86.46.8+1.0Moderate–High
Frontend / application development6.57.17.6+1.1High
Entry-level developer6.37.48.1+1.8High

Scores run 1–10, where 10 is most at risk. For context: the median profession in this edition sits around 5.5, bedside nursing scores 2.8, a service electrician 2.5, and licensed engineering 4.0.

Read the table this way: nothing in software scores below 4.7, and the entry rung is the second-highest score in the entire index. The field is protected at the top by judgment and accountability, and by nothing at all at the bottom.


Why computer science scores the way it does — the six factors

Here's how an entry-level developer answers them.

How much of this job can AI already do? (35% of the score)

More than any other professional work we score, and by a distance.

Writing functions from a specification. Boilerplate and scaffolding. Unit tests. Standard CRUD applications. Bug fixing on well-understood code. Documentation. Code review of routine changes.

That is a substantially complete description of a junior developer's week in 2019.

What resists: architectural judgment, debugging something nobody has seen before, deciding what to build, and owning a system when it fails at 3am.

How hard will it be to land that first job? (15%)

The worst entry-path erosion we measure anywhere in this index.

Stanford's Digital Economy Lab, working with ADP payroll data covering millions of workers, found that entry-level software engineering postings in the United States fell 67% between 2023 and 2024. That's payroll data, not a survey.

The share of tech jobs requiring three years of experience or less fell from 43% in 2018 to 28% in 2024. Big Tech new-graduate hiring is down more than 50% from 2019 levels, and new graduates now account for roughly 7% of Big Tech hires.

Does it have to be done in person, with your hands? (15%)

None of it. Zero, permanently.

Software is text in, text out — which is both why it can be done from anywhere and why nothing physical protects it.

Does someone need a human they can trust and hold responsible? (15%)

At senior level yes; at entry level almost not at all.

A principal engineer is trusted with architectural decisions and answers for them. A junior developer produces code that someone else reviews, merges and owns.

Does the law require a licensed human? (10%)

No. Nothing anywhere.

There is no software licence, no professional registration, no reserved activity. Compare licensed engineering, where the PE stamp is a legal monopoly, or nursing, where scope of practice is written into statute.

Two of our six factors sit at effectively zero for everyone in software, at every level of seniority, permanently. No individual action changes that.

How often does the job hit genuinely new, high-stakes situations? (10%)

At senior level frequently; at junior level rarely.

Designing a system that will handle unpredictable load is novel work. Implementing a well-specified feature is not.


The contradiction — and how to read it

This is the section that matters most, because the two headline numbers point in opposite directions and most coverage picks one.

The pessimistic number: 6.1% unemployment among recent CS graduates, 7.5% among computer engineering graduates — both above the 5.7% average across all recent graduates, and well above several humanities majors.

The optimistic number: 15% projected growth in software developer employment through 2034, with around 129,200 openings annually across the whole economy.

Both are true, and they describe different populations.

The profession is growing. The entry point to the profession has contracted sharply. When experienced developers say they still get recruiter messages, and new graduates say they've sent two hundred applications, they are both describing the market accurately — just not the same market.

Four forces stacked at once, and only one of them is AI:

  1. A tech-layoff overhang of 500,000-plus experienced engineers entering the market
  2. Return-to-office mandates narrowing the geographic funnel
  3. Offshoring of entry-level work
  4. AI raising the bar for what a junior hire needs to be able to do

We score AI exposure, and we'd be overstating our own framework if we let it take credit for all of this. A meaningful share of the entry-level collapse is cyclical and market-driven. What our score captures is that the content of junior work — not just its quantity — is what changed.

And the pay is still there, which is why students keep choosing it. New York Fed data puts CS early-career median pay at $87,000 and mid-career at $120,000, ahead of mathematics and most engineering disciplines. The upside didn't move. The entrance did.


What the trend shows

Entry-level developer: 6.3 → 7.4 → 8.1. Up 1.8 points in three years — the second-fastest movement in this entire index, behind only junior law.

Senior architect: 4.6 → 5.0 → 5.4. Up 0.8.

The gap widened from 1.7 points to 2.7. Software isn't sinking. It's splitting — and it split faster than any profession we measure except law.

What moved: automatability rose steeply, and entry-path erosion rose faster still. Both factors moved together, which is unusual and is why this score climbed so quickly.

What didn't move: physical and regulatory protection sat at effectively zero throughout. They were never there to lose.


How durable is the protection?

ProtectionStrength todayDurabilityWhere the pressure is coming from
Judgment under noveltyReal at senior levelErodingGenuine architectural novelty resists automation; routine design does not.
Human trust / accountabilityReal at senior levelDurableSomeone must own the system when it fails. Commercially rooted rather than legal.
Regulatory / licensingNonen/aNo licence exists. The exception is safety-critical work, which is why it scores best.
Physical / embodiedNonen/aSoftware has never had this and never will.

The honest read. Software has two of the four protections, both seniority-dependent, neither inherited from the profession. A doctor is protected on day one by a licence. A developer is protected only once they are trusted with something.

That is the core problem: the protection has to be earned, and the route to earning it is precisely what contracted.


How the role will actually change

Was the jobIs becoming the job
Writing the functionSpecifying it, then reviewing what was generated
Learning by writing bad code and being correctedLearning by reviewing generated code — which teaches differently
Implementing a well-defined ticketDeciding whether the ticket is the right thing to build
Debugging your own mistakesDebugging mistakes you didn't make and can't remember making
Junior → mid → senior over yearsFewer junior seats, higher bar for the first one

The row that should worry the profession is the second. Generations of engineers built judgment by writing something badly, watching it break, and fixing it. Reviewing generated code may build judgment too — but nobody knows yet whether it builds it as well, and the industry is running the experiment on an entire cohort without a control group.


Human strengths that will matter most

  1. Debugging genuine novelty — the failure nobody has seen, with no matching pattern. The most valuable and least automatable skill in software.
  2. Architectural judgment — deciding how a system should be shaped, and what it must not do.
  3. Accountability for a running system — being the person who answers when it breaks.
  4. Deciding what to build — product judgment, which sits upstream of all the code.
  5. Verification — knowing when generated code is subtly wrong in ways that pass tests.

Notably absent: speed of implementation, breadth of syntax knowledge, and volume of code produced. These were the traditional markers of a strong junior developer and are the fastest-depreciating things on the list.


Other things to consider about this job

What's genuinely good about it

The work is genuinely excellent, and it's worth saying so given the tone of everything above. Software is one of the few fields where a person can build something real, alone, with no capital and no permission.

What's genuinely hard about it

Getting the first job, which is the dominant risk and covered above.

Continuous obsolescence. The specific tools change every few years. This is energising for some people and exhausting for others, and it never stops.

Interview processes are unusually punishing — multi-round technical interviews with a median search of seven to nine months and, by some estimates, 200–400 applications per offer in the current market.

The work is invisible when it goes well and highly visible when it doesn't.

Who this work suits — and who it doesn't

Software tends to suit people who:

It's harder for people who:

What else is moving this market

Supply is high and unconstrained. CS is among the most-taken degrees and nothing caps numbers. Combined with the entry contraction, that's a crowded funnel.

Experienced engineers are doing fine. The market is bifurcated, not depressed.

Demand is genuinely growing at the aggregate level — 15% projected growth is real, and it's spread across healthcare, logistics, finance, defence and traditional enterprises rather than concentrated in the companies everyone applies to. Those employers see far fewer applicants per opening.


The AI-native advantage — what to actually do about it

Here is the paradox worth naming, because it's the most counterintuitive finding in this index: being close to AI does not protect you from it. A data analyst is not protected by understanding AI, any more than a copywriter is protected by understanding language. Exposure is set by the shape of the work — structured inputs, text output, no physical presence, no licensed accountability — and software answers yes to all four.

So the advantage isn't in knowing about AI. It's in what a person does with that.

What an AI-native developer actually looks like

1. Verify. Knowing when generated code is subtly wrong in ways that compile and pass tests. This is the most valuable and least teachable skill in the field right now, and it only works on top of fundamentals you actually built yourself.

2. Specify. Framing a problem precisely enough that the output is usable. The skill has shifted upstream from implementation to specification.

3. Debug what you didn't write. Increasingly the default condition, and materially harder than debugging your own work.

4. Judge architecture. Deciding what a system should be, which is the thing generation cannot do because it requires knowing the constraints that aren't in the ticket.

5. Own something running. Everything in this profile points at accountability. Anything that puts a graduate closer to a system they answer for moves them up the scale.

Concrete preparation

Before the degree: build something real and ship it. Not a tutorial project — something with users, however few. In a field with no credential, demonstrable work is the credential, and almost nobody your age has any.

During: take systems, networks, operating systems and compilers seriously rather than treating them as hurdles. Those are the fundamentals that verification depends on. Get internships early; conversion rates are falling but they remain the single strongest signal.

Aim deliberately at the protected end: embedded and safety-critical, security, ML engineering, or software in a regulated domain. These score between 4.7 and 6.0 rather than 8.1, and they're where the accountability lives.


The routes in

RouteNotes
CS degree at a strong programStill the main route. Fundamentals matter more than framework coverage.
CS with a domain — health, energy, finance, defenceThe domain is the protection. These employers see far fewer applicants.
Software engineering apprenticeshipExpanding, paid, and solves the experience problem directly. Seriously underused.
Embedded / electronics-adjacent routesScore best in the field, and consistently under-chosen.
BootcampsWeakest position in the current market — they optimise for the entry-level roles that contracted most.

Where a CS degree can take them later

Genuinely broad: software engineering, security, ML and data infrastructure, embedded and robotics, technical product management, developer relations, technical consulting, engineering leadership, and founding something.

The pattern is the same as everywhere in this index: roles where someone owns a running system score better than roles that produce components of one.


What to look for in a CS program

1. Fundamentals over frameworks. Systems, operating systems, networks, compilers, algorithms. Frameworks change; fundamentals are what verification rests on.

2. Real internship or co-op placement — and the conversion rate. Ask what proportion convert to full-time offers. Nationally that figure has fallen to around 53% from nearly 58%.

3. Does the curriculum teach working with AI tools — and checking them? A program banning them is training for a job that no longer exists. One teaching them uncritically is worse.

4. First-destination outcomes with job titles, not employment rates.

5. Specialization pathways in security, embedded, or ML — the protected end of the field.


Questions to ask a CS program

On outcomes:

On placement:

On the curriculum:

Red flags: outcomes quoted only as employment rate; a curriculum organised around currently-fashionable frameworks; no internship provision; evasiveness about conversion rates.


Go deeper — further reading

The counterargument — read this one:


Bottom line

Computer science is not a bad choice. It is a bad default.

For twenty years it was the safe recommendation you could make to a teenager without knowing anything about them. That is what has stopped being true — not because the field is dying, but because the gap between its best and worst outcomes has widened faster than almost anything else we measure.

The protected end is real — embedded and safety-critical at 4.7, security at 5.3, senior architecture at 5.4, ML engineering at 6.0. Good work, well paid, genuinely durable.

The entry rung is the problem, and it's severe. Entry-level postings down 67% on payroll data. Big Tech graduate hiring down more than half. The share of roles open to under-three- years experience falling from 43% to 28%.

So the useful advice isn't "don't do CS." It's: aim deliberately at the protected end from the start rather than assuming you'll drift there, take fundamentals seriously because verification depends on them, build something real with users before you apply anywhere, and look hard at apprenticeships and domain-paired routes.

And it's worth saying what the risk score can't measure. Software is one of very few careers where someone can build a working thing, alone, with nothing but a laptop and time — and where nobody's permission is required to try. That is rare, and it's still true.



Discussion questions

Part 1 — About the work itself

  1. What made computer science appealing? Push past "I'm good at computers" to something specific.
  2. Do they enjoy the debugging, or only the building? It's the honest test — most of the job is finding out why something doesn't work.
  3. Have they built anything that other people actually used? What happened?
  4. What does a Tuesday look like in the job they're imagining?

Part 2 — Reacting to the findings

  1. CS graduate unemployment is 6.1%; art history is 3.0%. Was that surprising?
  2. Entry-level developer scores 8.1 and senior architect 5.4. Does that change where they'd aim?
  3. The profile says CS is "not a bad choice but a bad default." Does that land, or does it feel unfair?
  4. Two of the six protection factors sit at zero for everyone in software, permanently. What do they make of that?

Part 2b — The harder conversation

  1. What's the plan if they don't get a software role at six months? Have you discussed it?
  2. Entry-level postings fell 67% on payroll data. Do they think that recovers, and what's their reasoning?
  3. Have they looked at first-destination job titles from their shortlisted programs?

Part 2c — The other side

  1. Of what's good about software — creation, fast feedback, pay, portability, no gatekeeper — which two matter most?
  2. Looking at "who this work suits": which items sound like them? Answer separately, then compare.
  3. Can they be stuck for a whole day without losing the thread? Think of a real example.

Part 2d — The AI question, practically

  1. The profile argues being close to AI doesn't protect you from it. Does that make sense to them?
  2. Of the five capabilities — verify, specify, debug what you didn't write, judge architecture, own something running — which appeals most?
  3. If they're already using AI to write code, has it made them better or just faster? Honest answer.

Part 3 — Testing it against reality

  1. The most valuable single action: find someone 2–3 years into a software career and ask what they actually did in year one, and whether that job still exists in the same form.
  2. Can they ship something small with real users before applying anywhere? In a field with no credential, evidence is the credential.

Part 4 — About the program

  1. Ask each program the outcome and internship-conversion questions from section 14.
  2. How much of the degree is systems-level rather than application-level?
  3. Are there specialization pathways in security, embedded or ML?

Part 5 — Widening the frame

  1. If CS weren't available, what would they choose?
  2. Have they looked at licensed engineering? Same technical aptitude, pointed at the physical world — 4.0 for site-based work against 8.1 for entry-level software. It's the most decision-relevant comparison in this index for a mathematically-minded student.
  3. Do they want software, or what software is made of — building things, solving puzzles, technical mastery, or the salary? Those lead to genuinely different places.

Part 6 — For the parent, alone

  1. Did I encourage this? If so, on what evidence — and is that evidence still current?
  2. "Learn to code" was good advice for twenty years. Am I still giving it out of habit?
  3. What's the one thing I most want them to understand — in a sentence, without a number in it?

This is analysis and informed judgment about a fast-moving situation, not a guarantee or a prediction. For computer science specifically, we've flagged that a meaningful share of the entry-level contraction is cyclical and market-driven rather than caused by AI — and that our framework would be overstating itself to claim otherwise.

Scores for 2023 and 2025 are retrospectively calculated using the current methodology. Scores measure exposure to what AI can already do — not how much any particular employer has deployed.


This profile is free. The other twenty-six are not.

If it was useful, the full profiles for nursing, medicine, law, business and psychology are the same length and standard — and cover the careers most families are weighing against this one. 1 profile $19 · 3 for $29 · 5 for $39.

28 careers, scored the same way. Scores measure exposure to what AI can already do — not how much any particular employer has deployed.
2023 and 2025 figures are reconstructed using current methodology, not archived from past editions.
Re-scored every six months. We publish where we might be wrong.
Analysis and scoring judgments are ours. Drafting is AI-assisted — how this is written.
Terms · Privacy · Refunds