Is Theoretical Computer Science Dead? What the Record Shows

The short answer

Nothing we found shows that theoretical computer science is dead, and nobody can honestly promise what it will look like in ten years. Job posts on CSTheory Jobs, a voluntary board maintained by the field's CATCS committee, are below their 2022 peak but within the range seen since 2018. Named researchers disagree, and some of them, including Grochow and Meir, say nobody can know yet. The changes we can document directly are in what conferences require. How credit will work is still being argued over, and the same record includes layoffs and proposed funding cuts whose causes nobody has separated. This page collects that record so you can judge for yourself. It is not advice and it promises no outcome.

If you came here for programming jobs in general, this page won't help. It covers academic theory: complexity, algorithms and the people who prove things about them. For the results behind the worry, see the ledger.

What the job posts show

CSTheory Jobs is a board where employers post openings in the field. We counted its posts from January through October of each year through its public API. The AI column counts titles containing AI, ML, machine learning or artificial intelligence.

Read this with care:

• It is a small voluntary sample, not a count of theory jobs, and it can't show that AI caused anything.
• Posts fell each year from 2022 to 2025, from 117 to 65, then rose to 76 in 2026. That is above 2018, 2019 and 2025, equal to 2024, and below 2020 through 2023. Across the nine years the series does not fall steadily.
• 2026's 8 AI titles are the most of the nine years, but the counts are small and noisy. Four of 2020's 7 came from one University of Copenhagen hiring round. In 2026, two titles match only on an institution's name and two come from one institute. Counting each institution once and dropping name-only matches gives 4 for 2020 and 5 for 2026. We did not apply the name filter to the other years.
• The 2026 figure was read on November 1, 2026, and posts can change.

Why people are worried

• Credit may move away from checking and explaining. Impagliazzo, Paturi and Williams (October 26) warn that recognition could go to whoever publishes first or has the strongest tools, while checking and explaining earn less. They also ask how to support students through the change. They are organizing the examination of OpenAI's ETH claim, which TechCrunch says OpenAI asked them to help interpret, and they say that role is not certification.
• Long projects may look too risky. Dor Minzer told Quanta (October 7) that AI removes the lesson of understanding failure, and he worries that fear of being scooped by a large company will deter ambitious work.
• Fewer theorists may be needed. Or Meir (October 14, edited October 20) doubts any career is safe from AI in ten years. He adds that falling demand for CS degrees could cut academic jobs, and that productive human-AI work, though maybe more fun, might reduce demand for theorists.
• Old reassurance is harder to give. Lance Fortnow (August 19) says his line to PhD students about rivals' limited bandwidth is hard to make against AI, though he still tells those who love math to continue and to use AI to go further.
• Advisors may change. Scott Aaronson (October 7) says some faculty might grow reluctant to take students, more immediately because of NSF cuts, while others, he guesses, will train students through heavy AI use plus traditional learning.

Arguments that the field will continue

• Teaching pays the bills. Joshua Grochow (October 12), who says he is active in the Association for Human Mathematics, calls the career still viable while admitting nobody can be sure. He notes that most academic theorists are paid mainly to teach, and reads the weak US job market as cyclical.
• Harder questions open up. TechCrunch (October 27) reports Ryan Williams as optimistic but worried about training future scientists. He says AI pushes researchers toward harder questions once the old ones become easy.
• The work could shift toward ideas. Kothari, Meka and Raghavendra (September 19), speaking only for themselves, propose a conference track judged on new definitions, questions and theories, with Lean certificates in place of written proofs. Asking readers to assume AI provers that settle well-specified claims within hours, they believe people will stay central to definitions, questions and theories, and they concede human costs the proposal doesn't address. Their post is about research, not jobs.
• Curiosity-driven work may remain. Boaz Barak (August 24), who works at both Harvard and OpenAI and says his opinions are his own, shares both excitement and grief. He expects education and curiosity-driven work to keep human mathematicians needed.

Emanuele Viola (September 9) sits between these camps. He calls AI hugely exciting and the disruption scary, and hopes it ends rewards for technical weight-lifting.

Several of these views predate OpenAI's October 6 release and the ETH claims.

Two pseudonymous answers on the Stack Exchange thread take opposite lines. One (October 13) expects no automation of core theory work and calls the boom a fad. Another (October 23) says nobody knows how hiring committees will judge AI-assisted records and advises against going all in on pure theory.

Pressures beyond AI

Not all of this is about AI, and no source we found separates the causes with numbers.

• Fortnow (August 12) reports losing his Illinois Tech post along with about 160 colleagues after a financial-exigency declaration. He attributes it to falling foreign graduate enrollment and several wider pressures, AI among them.
• Peter Brass (April 14), a former NSF theory director giving his own reading, says the FY2027 White House request cuts NSF's computing directorate from $0.9B to $0.3B, and that NSF projects its funding rate for all research proposals falling from 18% to 7%. He notes that Congress restored much of a similar cut last year. His post doesn't discuss AI.
• In 2025, Fortnow forecast a rocky theory faculty market, citing AI, university finances, and US visa and funding policy.

Rules and calls that have changed

• FOCS 2026: only humans may be authors, substantive LLM use should be disclosed at the end of the paper, and authors stay fully accountable.
• STOC 2027: experimental rules include a five-submission cap, public posting of all submissions and a required explanatory video, with exceptions by request to the chair. The committee says it will weigh exposition and proof clarity heavily and won't weigh AI use in evaluating papers.
• ECCC: warns that submissions with an essential part apparently AI-generated, or that humans would struggle to follow, can be turned down whatever their merit.

The next three address mathematics and are voluntary pledges, requests or funding, not rules for theory venues.

• Association for Human Mathematics: members pledge no labor or publicity for commercial AI firms and no AI-generated publications. Its Teaching Working Group urges that the vast majority of mathematics doctoral work avoid generative AI and that students and advisors put any AI use in a written agreement. That report is a working-group view, not the whole membership's.
• AGMAI, an advisory group that names the Institute for Advanced Study on its site, asks labs to stop testing advanced problems on proprietary models and to fund community-led understanding of what they release. It says it came together after OpenAI approached some members about an advisory board, and chose to be independent instead.
• The Simons Foundation (October 15) announced $15 million in combined funding and resources from itself, Anthropic and OpenAI, for workshops and initiatives that help researchers evaluate AI-assisted results.

What nobody can measure yet

These are the gaps we looked for and couldn't fill. Our searches weren't exhaustive.

• Theory PhD admissions. The CRA Taulbee survey doesn't split admissions by research area. In a fixed group of departments tracked every year, new CS doctoral students were 1,227, 1,714, 1,350, 1,612, 1,701 and 1,446 for 2020 to 2025. Across all 128 units reporting in 2025, the figure was 2,803. CRA describes enrollment as cooling, down 15% from 2024 and 15.6% from the 2021 peak, though still 17.8% above 2020. None of this is theory data. Among 2025 computing PhDs, it lists 76 with theory and algorithms as their primary area, with many areas unreported.
• Theory funding and tenure-track openings. We found no yearly series for theory funding as a whole or for tenure-track openings. One rough proxy: NSF awards under programs named Algorithmic Foundations numbered 90, 76, 81, 49 and 38 for 2022 to 2026, with 2026 running through September. That is one program at one funder, shaped by NSF's budget and delays, and it says nothing about AI's role.
• How hiring committees will treat AI-assisted work. We found no written standard. Conference rules cover submissions, not hiring.
• Cause and effect. We found nothing that separates AI from budgets, visas, demographics and university finances.
• Missing voices. We found no named PhD student or postdoc speaking individually about theory careers. The AHM report lists doctoral students and a postdoc among its authors, but it is a group document on mathematics. We found no lab statement on theory careers specifically. OpenAI's repository README says it leaves matters of credit to the math community, and we could not open three openai.com posts, so they are not reflected here. Named forecasts of the job market come mainly from Fortnow. Most of the voices here are US-based, so treat this as a partial picture.

Questions to ask before committing

These are our suggestions, drawn from the gaps above, not findings.

• Ask a prospective advisor what they expect of students on AI use, and whether it would be agreed in writing.
• Ask recent graduates of the group where they went, since the public numbers won't tell you.
• Ask how the department would judge AI-assisted work at hiring and tenure. If they don't know, that is an honest answer and worth knowing.
• Ask what you would do with the training if academic jobs don't appear.

Source: the linked posts, papers, conference pages and APIs, checked on November 1, 2026.

Evidence boundary: We read these public sources and paraphrased them. We did not interview anyone, and some affiliations are as stated on each person's own page. Claude, which helped draft and check this page, is made by Anthropic, a party to some of the events described and a funder named above. See the method.