Example · Computer Science Student

50% of the work a computer science student is heading toward still needs a person.

The first job is harder to land than it used to be. Your degree still leads to well-paid work.

Make my planA typical week at the middle pace. Your plan uses your own tasks.

Is computer science still worth it with AI?

In a typical week at the middle pace, 25% of this work needs your judgment and 25% is work you do using AI. AI could do about 50%. The numbers on your degree show a hard start and a strong payoff at the same time. Recent computer science graduates had a 7% unemployment rate in 2024, one of the highest of any major, and the same cohort had close to the highest starting pay of any major (New York Fed, The Labor Market for Recent College Graduates, February 2026).

Read more

Read together, they say the field still pays well, but getting in is harder. A model can now do a large share of what a first-year engineer was hired to do, and the people who land a first job are increasingly the ones who arrived with evidence of the other share.

That other share includes turning a vague request ("printer broken") into what the person actually needs, knowing whether a change is safe before anyone finds out the hard way, and being the person a system's users complain to. Many CS students already do some of this, at a part-time IT job, in a club or for a small tool other people use, and often describe it as the best part of their week. The semesters you have left are about making that visible to someone who hires, and about getting one paid summer before you graduate. Nothing in this report asks you to change major, and nothing in it asks you to become an AI researcher.

The week, task by task

Who does each hour of a typical 40-hour week.

See every task
You25% · 10 hours

Needs your judgment.

  • Standup, planning, retro, demos8%
  • Turning a vague request into a spec with the person who asked7%
  • Desk-side and in-person support, hardware, "says X, means Y"5%
  • Changes to live data, permissions, production config4%
You using AI25% · 10 hours

AI does the slow parts. You still own it.

  • Debugging failures with an unclear cause, in production or the lab9%
  • Environments, CI, dependencies, deploys, configuration7%
  • Reviewing others' changes, increasingly agent-written5%
  • Alert triage, monitoring, first response (ops and SOC rungs)4%
AI50% · 20 hours

AI could do most of it.

  • Feature tickets with a clear spec in an existing codebase19%
  • Bug fixes with reproduction steps10%
  • Writing tests and test cases for a change7%
  • Scripts, data pulls, one-off SQL, reports, dashboards (analyst rung)6%
  • + 2 smaller tasks9%
Why some tasks still need a person
Reading the ticket that says "printer broken" for what it means

The gap between what a user names and what they need lives in a building, a department's habits and a person's vocabulary. A model reads the ticket, but it does not know that the third floor calls the shared drive "the printer." Every task that holds up in this role has this shape.

Being the person users complain to when something you built breaks

If you have users, even a few, they know your name and come to you. A new graduate almost never has this. It is the single most convincing thing a student can show a hiring manager, and the plan is built around it.

Desk-side visits, lab imaging, hands on hardware

Someone has to be in the room. Nobody hires an engineer for it, but it is why IT support desks, including many on campuses, still hire students.

Debugging other people's broken code

As writing code gets cheap, the person who can tell whether it is right becomes the bottleneck. This is the skill the plan asks you to name and log.

Knowing what the machine is actually doing

Systems knowledge is how you tell that an agent's diff that looks fine will page someone at 2 a.m. It is also the track (operating systems, networks, databases) where entry hiring is holding up best among the jobs we studied.

Good to know

What is hard about this field, and what you can do about it.

The hard truth

Recent CS grads face higher unemployment than most majors.

A CS degree is no longer a quick ticket to a first job. Recent CS grads are now more likely to be unemployed than the average recent college grad.

What you can doStart the search early. Get an internship, ship real projects people use, and look beyond tech firms to banks, hospitals, utilities and government.
Sources (2)

What can I do with a CS degree if entry-level coding jobs shrink?

Plan A is the best move in the field you are studying for.

APlan A

Focus your studies

The main focus. Most weekly sessions build toward it.

Your next stepAim for infrastructure and internal-tools teams
Each week
4 sessions
Ready in
About 3 months to a supervised change
Why it fits

Why you. If you like knowing what the machine is doing or prefer debugging to writing, this is your door. Time near a campus department, lab or club that runs real systems helps if you can get it.

The gap. No production change on your record, no operating systems or networks course yet, and a resume that doesn't yet show the systems work you've done.

The trade-off. It is less glamorous. Pay on an infrastructure team at a hospital, utility, state agency or university is usually below a big-tech feature team, so compare offers. Some jobs are titled "support engineer," which you may need to explain to family. The door is much wider.

2 other moves that fit
  • Build tools for people you can talk to: internal tools, implementation and solutions rolesReady in about 3 months to the project
  • Still apply to big-company internship programs, without spending much on itReady in about a month to apply

What's a plan B for CS majors?

Plan B is meant to be different from this job. It was picked because skills from this role carry over.

BPlan B

Get ready for a new kind of job

A second option. A newer job AI is creating that these skills already fit.

A job to aim forAI deployment strategist

Solutions and implementation roles want juniors who talk with users, and reading a vague ticket for what it means is that skill.

Each week
1 session
Outlook
Hiring now
Pay
$100K to $290K a yearRange across five US postings in 2026, latest Oct 2026; some require Python.
2 other options
  • AI workflow builderDebugging other people's broken code means you can ship and maintain working automations for a team within a few months.
  • AI evaluation specialistKnowing what the machine is actually doing helps you design tests that catch where an AI coding tool fails without warning.

In your own plan you can switch to any of the 47 roles.

The 13 weeks

Small steps that add up. One plan for both: most sessions build Plan A, one a week builds Plan B.

By week 13 you will have

AFor Plan A

  • A project with real users
  • A changelog
  • An error log of agent mistakes you caught
  • About 30 summer applications sent with a one-page summary of your work

BAnd for Plan B

  • A launch plan and a feedback memo
  • An AI deployment playbook for your portfolio
Month 1Weeks 1 to 4
  1. 1See where your time goes
  2. 2Pick a project with users
  3. 3Find employers outside big tech
  4. 4Start your error log

You finish withFive review comments on your own commits

Month 2Weeks 5 to 8
  1. 5Ask to shadow a real change
  2. 6Build one feature for a user
  3. 7Tests, spring classes and practice
  4. 8Finals week, count your catches

You finish withA count of agent errors you caught, by type

Month 3Weeks 9 to 13
  1. 9Finals week, plan the change
  2. 10Break: do the change
  3. 11Break: find paid summer options
  4. 12Break: make one open-source fix
  5. 13Break: send your applications

You finish withFive more applications and a paragraph choosing your spring focus

Three courses from the plan
  • GitHub Student Developer PackGitHub Education · Free to verified students

    Yes, Week 1.

  • The Missing Semester of Your CS EducationMIT CSAIL · Free

    Yes, Weeks 1, 2, 5.

  • How to Do a Code ReviewGoogle (eng-practices) · Free, open source

    Yes, Week 4.

Checked 2026-09-22.

The fuller picture

Risks beyond AI: moderateHiring-cycle risk

Tech headcount moves for reasons unrelated to AI: the 2021–22 hiring surge, the 2022–23 correction, interest rates, and a federal tax rule that from 2022 through 2024 made companies spread deductions for software development pay over five years, now reversed. The supply side is turning in your favor. CS enrollments have started to fall, so a later cohort graduates behind the wave.

Where this could be wrong
  • The destination blend is ours. We assumed half of graduates land on software teams and a fifth in IT operations. Your school's placement data would correct it, and the career center has it. If you lean toward IT operations, the share of your week that holds up rises and starting pay falls. Feature teams work the other way.
  • The Taulbee cohort is not the country. Its 32,266 degrees come from doctoral-granting departments; NCES, the federal education statistics agency, counted 108,503 CS bachelor's degrees nationally in 2021–22, the latest year we could reach. The enrollment drop may be larger or smaller at schools like yours.
  • Three research groups disagree on how much of this is AI. Anthropic calls its hiring result barely significant; Stanford calls its 19% descriptive; the New York Fed's regional firms report retraining, not cuts. We took the middle. The direction is the same either way, though the urgency differs.
  • We may be leaning too hard on campus systems work. Not every student can get near it. If that door is closed to you by Week 5, the user-facing move moves to first and your project carries the plan.

This is an example. Your plan uses your own week.

Make my plan

Also matched: computer science student, cs student, cs major. Research checked 2026-09-22. Page updated .