Home · Business · IT, Data & Security · Software development

YES

As of 13 August 2026, AI can fix your broken code.

This still needs a person who signs their name to it.

Can you do it?

5 minutesto a draft.

30 minutesto something you’d act on.

Cost, all in£0

Skill neededchat-fluent

Who has to check ityou

What the alternative costsA professional developer remains the alternative; the supplied tool data gives no price for that service.

If this goes wrong: the patch can hide the original defect, introduce a security problem or take a working service offline.

What to actually do

  1. Hand it to a person

    The route this page recommends

    A person who owns the outcome does this end to end, worth it when the failure is dear.

  2. Use a tool built for this

    Second choice

  3. Do it yourself

    The distant third

    A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

    How to actually do it

    1. Open the project in version control, create a new branch, and make sure you can return to the last known working commit.
    2. Reproduce the failure locally and gather the exact error message, stack trace, command used, expected result, actual result and the versions of the language, framework and dependencies.
    3. Remove passwords, API keys, tokens, private customer data and other secrets, then paste the relevant code, configuration excerpts and tests into the prompt.
    4. Ask the model to diagnose the evidence before proposing a patch, and require it to state assumptions, suggest the smallest change and add a test for the failure.
    5. Apply the proposed change on the branch, run the model's commands, run the existing test suite and compare the new behaviour with the expected behaviour.
    6. Inspect the diff for changes to authentication, authorisation, input validation, data handling, dependencies and production configuration before asking a colleague or developer to check any high-impact change.
    7. Deploy only through your normal review and release process, then monitor the affected feature and keep the rollback commit available.

    Prompt

    Help me diagnose and fix this software bug. Do not invent files, APIs, dependencies or test results. First state what you think is happening and list the evidence for that diagnosis. Then propose the smallest safe change, showing the complete replacement code only for the files or functions that need changing. Explain each change and identify assumptions. Add or update tests that reproduce the bug and prove the fix. Tell me the exact commands to run, what output I should expect, and what would disprove your diagnosis. Do not suggest deleting data, disabling authentication, weakening validation, exposing secrets or changing production systems directly. If important information is missing, ask focused questions instead of guessing. Here is the context: language and version: [LANGUAGE_AND_VERSION]. Framework and dependency versions: [VERSIONS]. Operating system and runtime: [ENVIRONMENT]. Expected behaviour: [EXPECTED_BEHAVIOUR]. Actual behaviour: [ACTUAL_BEHAVIOUR]. Exact error message and stack trace: [ERROR]. Steps to reproduce: [STEPS]. Relevant code and tests: [CODE_AND_TESTS]. Recent changes: [RECENT_CHANGES]. I have removed passwords, API keys, tokens, customer data and other secrets.

    Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

What it gets wrong

  • AI cannot see the full runtime, deployment environment or operational history unless you provide them.
  • AI can recommend a patch that passes a narrow test while leaving a wider bug or security flaw in place.
  • AI cannot decide whether a behaviour change is acceptable for your users, contracts, data or service commitments.
  • AI cannot take responsibility for an outage, corrupted data or a vulnerability caused by its suggestion.
  • AI cannot safely infer missing requirements from a vague description of what the software should do.

Even on a YES, the friction has a name: judgement under ambiguity, verification cost and stakes of error.

How we scored this

Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.

AxisScore (0–2)
Output2
Inputs2
Verification1
Liability1
Effort delta2
Total8 / 10

The methodology and its thresholds are published in full.

FAQ

Can ChatGPT fix my broken code?
Yes, for many ordinary bugs it can diagnose the error and draft a patch, especially when you provide reproducible steps, the exact error and relevant code. Run the change in a branch, test it, and do not paste secrets or private customer data.
Is AI safe to use for debugging code?
It is useful for diagnosis and test-writing, but a plausible fix can still introduce a security defect or break another part of the system. Keep the change out of production until you have run the relevant tests and checked the diff.
Can AI fix code without seeing the whole project?
Sometimes it can fix a self-contained error from a small code sample and its test. For bugs involving configuration, dependencies, state, deployment or interactions between components, missing project context makes confident answers unreliable.
Should I let AI deploy the fix to production?
Not without your normal review, testing, backup and rollback process. AI can prepare commands or a pull request, but you or a suitably skilled colleague must decide whether the change is safe to release.

Nearby answers

All Software development rulings

Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.

The newsletter

AI news, new answers and product picks, straight to your inbox.