1. Can AI do a task?
Example: search for a formal proof in a defined setting and let a proof assistant check it.
This is the strongest kind of evidence on this site.Career outlook · Evidence first
AI and TCS careers: what changes, and what does not
AI can change parts of a research or engineering job without telling us how many jobs will exist in the future. This page separates what has been shown, what is still debated and what is only a forecast.
Important boundary:
This is not a salary guide or a job-market forecast. The sources checked for
this site do not support a reliable claim about broad job losses, job gains
or future pay. We will not invent those numbers.
Do not skip a step
There are three different questions.
A real task-level result does not automatically answer a career-level question. Keeping the layers separate makes the evidence easier to judge.
2. How will teams change?
People may spend less time on some checks and more time framing problems, reviewing output or working across fields.
The direction is plausible, but broad workplace effects need evidence beyond a few tool demonstrations.3. How many jobs will exist?
Forecasts about net job numbers, salaries or the disappearance of a profession require labour-market evidence.
A single AI result is not that evidence.What to learn
Practical guidance for three starting points.
This is decision guidance, not a prediction about the job market. The aim is to build skills that help you check whether an AI result is useful.
Learn the fundamentals before the shortcuts.
Use AI to explain unfamiliar ideas, but write out the definitions and reproduce the proof or algorithm yourself.
- Practise reading precise statements.
- Learn to test whether an answer is correct.
- Keep building mathematics and programming foundations.
Treat AI as a research assistant, not an authority.
Try it on bounded search, formalisation and candidate generation, then keep the usual checks for proof and scope.
- State the problem and its assumptions clearly.
- Ask what checked the result.
- Separate a finite construction from a universal theorem.
Separate theory, benchmarks and deployment.
An AI-suggested algorithm needs a correctness argument, a fair baseline and evidence that it works under real constraints.
- Learn asymptotic reasoning and empirical testing.
- Check hardware, data and input-size assumptions.
- Look for the formal statement behind the demo.
Skills that keep their value
The ability to check is becoming more visible.
This is practical guidance, not a claim backed by a labour-market study. These skills directly help with the evidence problems shown on this site.
- Problem framing. A clear question is often harder than the generated answer.
- Mathematical reasoning. You need enough understanding to see when a confident explanation is empty or wrong.
- Formal verification. Proof assistants make the difference between “looks right” and “the stated proof checks”.
- Programming and experimentation. Real systems have limits that a clean proof or demo may hide.
- Clear communication. A useful result must be explained with its assumptions and limits intact.
Use the evidence