# Learn: how software gets maintained when AI does the work

Plain answers to the questions people ask about AI-written code, AI agents and software upkeep. Each page is written by the team that builds Bugstack, from what we have measured and what we run in production. Start with [what autonomous software maintenance is](/autonomous-software-maintenance).

## The category

-   [Can AI maintain software?](/learn/can-ai-maintain-software)  
    Yes, for the parts of maintenance that have a clear signal and a clear check. Which jobs AI can take on today, which it cannot, and what a person still does.
-   [Self-healing code: what the term means and what works today](/learn/self-healing-code)  
    Self-healing code is software that gets repaired without a person doing the repair. What the term promises, what works today, and why we call it maintenance.
-   [What is software maintenance? The four types, and which AI can do](/learn/what-is-software-maintenance)  
    Software maintenance is the work that keeps software running after it ships. The four types in plain words, why it is underfunded, and which parts AI can do.
-   [The AI on-call engineer: what it can take off your pager](/learn/ai-on-call-engineer)  
    What an AI on-call engineer does when something breaks at night, which pages it can handle without waking anyone, and which still need a person.
-   [A glossary of autonomous software maintenance](/learn/glossary)  
    Plain definitions of the terms used when AI agents maintain software: failing test, sandbox, CI, lockfile, autonomy level, revert, and more.

## What Bugstack does

-   [Automatic crash fixes: from a production error to a pull request](/learn/automatic-crash-fixes)  
    How Bugstack fixes a production crash on its own: it hears the error, reproduces it with a failing test, writes the fix and opens a pull request.
-   [Fixing a failing CI build automatically](/learn/fix-failing-ci)  
    When CI goes red on your main branch, Bugstack finds the change that broke it and opens a pull request with the fix. How it works and what it leaves alone.
-   [Fixing a failed deploy automatically](/learn/fix-failed-deploys)  
    When a deploy fails on Render, Bugstack reads the build log, finds the commit that broke it and opens a pull request that fixes it or restores the last one.
-   [Automatic dependency upgrades: packages kept current without the noise](/learn/automatic-dependency-upgrades)  
    Bugstack scans your packages every day and opens upgrade pull requests with the lockfile rebuilt. What it upgrades, what it holds back, and its limits.
-   [Automated vulnerability remediation: from an advisory to a patch](/learn/automated-vulnerability-remediation)  
    Bugstack checks your packages against published advisories every four hours and opens the upgrade that closes them. What it patches and where it stops.
-   [Secret scanning: finding keys and passwords committed to your code](/learn/secret-scanning)  
    Bugstack scans your repository every night for committed secrets and tells you exactly what to do about each one. Why it reports and never rotates.
-   [An AI security audit that proves what it finds](/learn/ai-security-audit)  
    A Bugstack security audit reads your repository, tries to prove each problem before reporting it, and starts the fixes. Your first audit is on us.

## Trusting AI-written code

-   [Is AI-generated code reliable?](/learn/is-ai-generated-code-reliable)  
    AI-written code is reliable when something checks it. What Bugstack measured on a public benchmark and a sealed exam, and what the numbers leave out.
-   [How to verify AI-generated code before it merges](/learn/verify-ai-generated-code)  
    Five checks that tell you whether AI-written code is safe to merge, in the order to run them, and how Bugstack runs the same checks on its own fixes.
-   [How to review a pull request an AI agent opened](/learn/review-ai-pull-request)  
    A short method for reviewing a pull request from an AI agent, what a good one contains, and what to send back instead of merging.
-   [Can AI fix bugs on its own?](/learn/can-ai-fix-bugs)  
    Yes, when the bug can be reproduced. How an AI system fixes a bug without being asked, how often it works, and which bugs still need a person.

## After you ship

-   [Your vibe-coded app is live. What breaks, and who maintains it?](/learn/vibe-coded-app-maintenance)  
    An app built with AI still needs upkeep once real people use it. What goes wrong first, what maintenance means in plain words, and who can do it.
-   [A security checklist for vibe-coded apps](/learn/vibe-coding-security-checklist)  
    Nine security checks for an app built with AI, most serious first, with what each means in plain words and which ones a tool can do for you.
-   [Vibe coding technical debt: what it is and how it gets paid down](/learn/vibe-coding-technical-debt)  
    Technical debt in an AI-built app is the upkeep you skipped to ship fast. The four kinds, how to tell how much you have, and what pays it down.
-   [Do you need a vibe coding cleanup specialist?](/learn/vibe-coding-cleanup)  
    When hiring someone to clean up an AI-built app is worth it, what they do, what it does not solve, and which parts of the job a tool can do instead.
-   [Who maintains code that AI wrote?](/learn/who-maintains-ai-written-code)  
    Whoever shipped it. AI wrote the code and the responsibility stayed with you. What that upkeep involves and the three ways to get it done.

## How agents work

-   [What is a sandbox for AI coding agents?](/learn/ai-agent-sandbox)  
    A sandbox is an isolated place where an AI agent can run code safely. What it should block, what it is for, and how Bugstack uses one to prove a fix.
-   [What is SWE-bench, and what is SWT-Bench?](/learn/what-is-swe-bench)  
    SWE-bench tests whether AI can fix real bugs. SWT-Bench tests whether it can write the test that proves a bug exists. What each measures, in plain words.
-   [How to read an AI agent benchmark](/learn/how-to-read-ai-agent-benchmarks)  
    Five questions that tell you whether an AI coding benchmark score means anything, with Bugstack’s own result used as the worked example.
-   [What is agentic coding, and what is a background agent?](/learn/agentic-coding)  
    Agentic coding is AI that carries out a coding task itself. A background agent does it without you watching. What each means and where maintenance fits.
-   [Autonomy levels for AI agents: suggest, review, autopilot](/learn/ai-agent-autonomy-levels)  
    How much an AI agent should do without asking, as three levels you can set per repository, and what should wait for a person at every level.
-   [Human in the loop or on the loop: where a person belongs](/learn/human-in-the-loop)  
    In the loop means a person approves each change. On the loop means a person watches and can step in. Which to use for an AI agent that changes your code.

## See it work on your own code.

$100 a month

Includes $100 of usage credit every month. Start with a 14-day trial: $50 of credit and a first security audit on us. A card is required.

[Get started](https://dashboard.bugstack.ai/login) [Book 15 minutes with the founder](https://cal.com/mason-bachmann-1atgdq/15min)