Will AI replace software engineers?
Not the profession, but much of the routine coding inside it. Software engineering scores 30.0/100 on the AI Resistance Index (Commoditized). Code is digital, cheap to generate and quick to test, which is exactly where AI is strongest. Deciding what to build, designing systems and answering for them in production still need people, and BLS still projects the occupation to grow.
Higher is safer. Scored on six structural dimensions, weighted as shown below.
How the score breaks down
Each dimension is rated 1 (AI holds every advantage) to 5 (strong structural protection). A rating's contribution is (rating − 1) ÷ 4 × weight, so a 1 contributes nothing and a 5 contributes the full weight.
| Dimension | Weight | Rating | Points |
|---|---|---|---|
| Automation Replaceability | 25% | 2 / 5 | 6.25 |
| Regulatory & Liability Friction | 20% | 2 / 5 | 5 |
| Human Judgment Density | 15% | 3 / 5 | 7.5 |
| Physical-World Coupling | 15% | 2 / 5 | 3.75 |
| Trust & Relationship Lock-In | 15% | 3 / 5 | 7.5 |
| AI Capital Intensity | 10% | 1 / 5 | 0 |
| Total | 100% | 30.0 |
Points are exact; the scoring engine rounds the total to one decimal.
Why each dimension scores the way it does
Rated 2. Writing functions from a clear specification, generating tests, refactoring, writing documentation and migrating old code are now everyday uses of AI coding assistants and agents. That covers a large share of the hours in a typical engineering week. It rates 2 rather than 1 because working out what users actually need and fitting new code into messy, real systems still takes a person.
Rated 2. Nobody needs a licence to write software, and a bug rarely carries personal legal liability for the engineer who wrote it. The exceptions are safety-critical and regulated domains such as medical devices, aviation and banking, where standards require documented human review and sign-off. Those pockets are real but cover a minority of the field.
Rated 3. Choosing an architecture, trading speed against reliability, deciding what not to build and judging security risk are real judgment calls with lasting consequences. But a large part of day-to-day implementation follows established patterns and libraries, and pattern-following is what AI models reproduce best, so judgment protects senior design work more than routine delivery.
Rated 2. Almost all of the work happens in editors, terminals and cloud consoles, and teams already work remotely across time zones. It rates 2 rather than 1 because some engineering is tied to hardware: embedded systems, vehicles, medical and industrial equipment, where code has to be tested against physical devices in a lab.
Rated 3. Teams lean heavily on engineers who know why a system was built the way it was, where it breaks and how to fix it at three in the morning. That institutional knowledge is hard to replace. But the trust attaches to particular people with long tenure, not to the role, and a new hire with good AI tools ramps up faster than ever.
Rated 1. The substitute is a coding assistant that costs a modest monthly subscription per seat and runs inside the tools engineers already use. A company needs no frontier-scale investment to adopt it, and every model release makes it more capable, so there is no capital barrier slowing adoption.
Which engineering tasks AI takes first
BLS lists the core duties of software developers and of quality assurance analysts and testers. Sorted by exposure:
Most exposed to AI
- Writing code for well-specified features and fixes
- Creating test plans and running tests, the core of QA work
- Documenting applications and systems for future maintenance
- Routine maintenance, upgrades and migrations of existing programs
Most durable
- Analysing users' needs and deciding what to build
- Designing how the pieces of a system work together
- Identifying project risks, including security, and deciding how to reduce them
- Owning reliability of systems in production and answering for failures
The labor market today
| Software developers: median pay (May 2025) | $135,980 |
| QA analysts and testers: median pay (May 2025) | $104,300 |
| Jobs, developers and testers combined (2025) | 1,905,400 |
| Projected growth, 2025–35 | +10% |
| Openings per year, on average | ≈106,100 |
| Typical entry-level education | Bachelor's degree |
BLS projects growth much faster than average and ties part of it to building software for artificial intelligence, the Internet of Things and robotics, along with rising spending on security software. Testers are paid noticeably less than developers, and their core duties sit closer to the exposed list above.
The score and the jobs outlook answer different questions. BLS estimates how many people employers will need; the AI Resistance Index measures how much of the work AI can do and how fast its price falls. Both can be true at once: more software gets built while routine coding gets cheaper, a squeeze that tends to show up first in entry-level hiring and rates rather than in total headcount.
How to raise your resistance
- Own systems, not tickets. Engineers judged on closing well-defined tickets compete directly with coding agents. Engineers who own a system end to end, including its uptime and its failures, hold the accountable part of the job.
- Move toward regulated or safety-critical work. Medical devices, aviation, payments and similar domains require documented human review. That review requirement is the closest thing software has to a licence.
- Get closer to hardware and infrastructure. Embedded, robotics, networking and on-premises systems carry physical constraints that slow full automation and keep hands-on debugging valuable.
- Become the one who directs the agents. Writing precise specifications, reviewing generated code critically and catching what the model got wrong is becoming its own skill. Treat AI output as a junior colleague's pull request, not finished work.
This score rates the occupation as a whole. Your exposure depends on what you actually do all day. The free Job Safety score asks about your specific role and day-to-day work, then scores you on four personal factors in about a minute.
Score my job — freeQuestions people ask
Will AI replace software engineers?
Not as a profession in the near term, and BLS still projects 10% growth from 2025 to 2035. What AI is replacing is a large share of routine coding, testing and documentation, which changes what employers pay for and how many junior engineers they need.
Why does software engineering score lower than plumbing?
The index rewards structural barriers to automation: licences, physical work, high-trust relationships and costly substitutes. Software has almost none of them. It is digital, unlicensed and cheap to automate, while plumbing happens inside buildings under licensing rules.
Is it still worth learning to code?
Yes, but as a means rather than the product. Coding knowledge is what lets you specify, review and fix what AI tools produce. The durable value is in system design, domain knowledge and owning results, so build those alongside the syntax.
Compare with
Sources and method
Ratings are Dawnstar Exploration's structural assessment using the six-dimension AI Resistance Index framework; labor-market figures are quoted from the sources above as published. This page describes an occupation, not any individual.