What AI coding assistants can and can't do, what studies say about productivity, the security risks, how the job of a developer is changing, and the skills worth building if you're learning to code now.
AI tools can now write working code from a plain-English request, explain unfamiliar codebases and fix bugs. That has led to bold claims that programming as a job is ending. The reality, so far, is more nuanced. Here's what the evidence shows.
What AI coding tools do well?
- Boilerplate and routine code: setting up projects, writing tests, converting data formats.
- Explaining code: summarising what an unfamiliar function does.
- Learning new languages and libraries: suggesting idiomatic patterns and examples.
- Small, well-defined tasks: a single function with clear inputs and outputs.
- Agents: newer tools can edit several files, run tests and iterate, under human supervision.
Where they struggle?
- Large, complex codebases with unwritten conventions and history.
- Ambiguous requirements: deciding what to build is often harder than building it.
- Confident mistakes: plausible but wrong code, invented functions or outdated APIs.
- Security and edge cases, unless specifically prompted and checked.
What the studies say about productivity?
Results vary depending on the task and the developer:
- A 2022 controlled experiment by GitHub researchers found developers using Copilot completed a specific task, writing a simple web server, about 55% faster than those without it.
- A 2025 randomised study by the research group METR found that experienced open-source developers working on their own large, mature projects took about 19% longer when allowed to use AI tools, even though they believed the tools had sped them up.
The takeaway: AI can speed up well-defined or unfamiliar tasks, but gains aren't automatic, and people aren't always good at judging them. Measure, rather than assume.
The security risks
- Insecure patterns: generated code can include vulnerabilities such as injection flaws or weak cryptography.
- Invented packages: models sometimes suggest software libraries that don't exist. Attackers have registered such names with malicious code, a risk sometimes called "slopsquatting". Always check a package is genuine before installing it.
- Leaking secrets: don't paste passwords, keys or confidential code into tools your organisation hasn't approved.
The UK's National Cyber Security Centre publishes guidance on secure development, which applies whether code is written by a person or a model.
How the job is changing?
Developers increasingly spend time specifying problems, reviewing and testing generated code, and designing systems, rather than typing every line. Skills that matter more than ever:
- understanding requirements and talking to users;
- reading and reviewing code critically;
- testing, debugging and security;
- system design and architecture;
- knowing when not to use AI.
Will there still be jobs?
Nobody knows for certain. Some routine coding work is likely to shrink, and entry-level roles may change most, which is a concern for people starting out. But demand for software keeps growing, and history suggests better tools often expand what gets built. The safest bet is to be a developer who uses AI well, not one who competes with it.
Learning to code now?
- Learn the fundamentals properly: you can't review code you don't understand.
- Use AI as a tutor: ask it to explain, not just to write.
- Build real projects and put them on a public portfolio.
- Practise debugging without AI sometimes, so you keep the skill.
- Contribute to open-source projects: reviewing other people's code and having yours reviewed teaches the judgement that AI tools can't supply.
- Learn version control with Git early; it makes experimenting with AI-generated changes much safer.
Sources and further reading
- METR: Early-2025 AI and experienced open-source developer productivity
- arXiv: The impact of AI on developer productivity (Peng et al.)
- NCSC: Secure development and deployment
Facts checked on 20 October 2026. Spotted an error? Tell us and we will correct it.
Frequently asked questions
What the studies say about productivity?
The takeaway: AI can speed up well-defined or unfamiliar tasks, but gains aren't automatic, and people aren't always good at judging them. Measure, rather than assume.
What should you know about the security risks?
The UK's National Cyber Security Centre publishes guidance on secure development, which applies whether code is written by a person or a model.
How the job is changing?
Developers increasingly spend time specifying problems, reviewing and testing generated code, and designing systems, rather than typing every line.
Will there still be jobs?
Nobody knows for certain. Some routine coding work is likely to shrink, and entry-level roles may change most, which is a concern for people starting out.
Related Articles
From Our Network
- Unlocking The Power Of Positive Vibes A Beginner S Guide To Optimising ↗
Chris is a freelance writer and editor covering a wide range of topics with a focus on accuracy and depth.
Last updated: 2026-04-25 · Fact-checked by editorial team