Tau
AI for software engineers · 11 lessons

AI for software engineers: ship better code safely

Use AI coding assistants to understand code, write tests, fix bugs and review changes, without letting errors slip through. For developers at every level and the people who lead them.

Who it’s for: Software engineers, from first job to staff and engineering managers.

AI coding assistants can write a function in seconds. Getting code you'd actually merge is harder. Without context about your conventions and architecture, an assistant guesses, and the guesses look plausible. In one 2025 study, experienced open-source developers were slower with AI on code they knew well, even though they felt faster. Knowing where it helps is a skill in itself.

This course starts there. You'll learn where AI speeds up engineering and where it slows you down, then write a project context file so its output fits your codebase. You'll trace a request through an unfamiliar repo and generate tests, then break the code on purpose to prove the tests fail when they should. You'll also debug a production error by running real experiments.

In later lessons, you'll work with an agent to take a ticket from spec to pull request in small, reviewable steps. You'll review AI-written code for logic, security and licence issues, and defend against prompt injection, leaked secrets and slopsquatting, where attackers publish packages under names that AI tends to invent. Finally, you'll measure whether AI is really improving your team's delivery.

By the end you’ll be able to…

  • Write specs and context files that get mergeable code from agents
  • Review AI-written pull requests for logic, security and licence issues
  • Defend against prompt injection and hallucinated dependencies
  • Measure AI's real effect on delivery with DORA-style metrics

Short on time? Start with the essentials

These 4 lessons cover the most useful ideas in this course. Each one takes about a minute.

Essential 1

Know where AI helps and hurts in the dev lifecycle

Identify which engineering tasks AI reliably speeds up and where it tends to slow experienced developers down.

Essential 2

Give the model the context a new teammate needs

Write a project context file covering conventions, architecture and commands so AI output fits your codebase.

Essential 3

Debug a production error with AI

Feed AI the stack trace, logs, recent diff and hypothesis, then use its suggestions to narrow the cause with real experiments.

Essential 4

Review AI-written code like a senior engineer

Apply a review checklist for AI code: logic, edge cases, security, performance, duplication and licence concerns.

Every lesson in this course

This course has 11 video lessons of about a minute each, from the basics to expert habits. Each one ends with a question, three takeaways and its sources.

Foundations 2 lessons

  1. Know where AI helps and hurts in the dev lifecycleEssential

    Identify which engineering tasks AI reliably speeds up and where it tends to slow experienced developers down.

    “In one trial, experienced devs were 19% slower with AI, yet felt 20% faster.”

  2. Give the model the context a new teammate needsEssential

    Write a project context file covering conventions, architecture and commands so AI output fits your codebase.

    “Same prompt, two outputs: one knows your repo's conventions, one is generic.”

Building skills 3 lessons

  1. Understand an unfamiliar codebase in an afternoon

    Use AI to trace a request through an unfamiliar repo, then confirm its explanation by reading and running the code.

    “Day one on a million-line repo? Ask AI to trace one request, then verify it.”

  2. Generate tests that actually catch bugs

    Prompt for behaviour-focused tests with edge cases, then mutate the code to prove the tests fail when they should.

    “100% coverage, zero assertions that matter. Here's how to get real tests from AI.”

  3. Debug a production error with AIEssential

    Feed AI the stack trace, logs, recent diff and hypothesis, then use its suggestions to narrow the cause with real experiments.

    “Paste the stack trace, logs and last diff, not just 'it's broken'.”

Advanced 5 lessons

  1. Take a ticket from spec to pull request with an agent

    Write a spec with acceptance criteria, let an agent plan and implement in small steps, and keep the diff reviewable.

    “Skip the spec and the agent hands you a 2,000-line diff nobody can review.”

  2. Review AI-written code like a senior engineerEssential

    Apply a review checklist for AI code: logic, edge cases, security, performance, duplication and licence concerns.

    “Only a third of developers trust AI code accuracy. Here's what to check first.”

  3. Stress-test a design doc with AI before review

    Have AI play a sceptical reviewer on your design doc, listing failure modes, scaling limits and missing alternatives, then decide which ones to address.

    “Ask AI to argue against your design doc before your staff engineer does.”

  4. Stop slopsquatting, secret leaks and prompt injection

    Recognise AI-specific security risks in your toolchain and apply controls: package verification, secret scanning and agent permissions.

    “Open-source models suggested non-existent packages about 1 in 5 times. Attackers notice.”

  5. Plan a large migration with AI in reviewable slices

    Break a framework or API migration into scripted, testable batches that AI agents can execute and humans can review.

    “Upgrading 400 files to a new framework version? Do it in slices you can review.”

Expert 1 lessons

  1. Prove whether AI is improving your team's delivery

    Set a baseline and compare lead time, change failure rate and review load before and after AI adoption, so decisions rest on data rather than impressions.

    “DORA found AI amplifies what's there: strong teams improve, struggling ones get worse.”

Apps we recommend

These are the tools this course uses, and what each one is best for. We have no partnership with any of them.

GitHub Copilot

Best for: In-editor completions, chat and PR agent inside GitHub

The most widely deployed coding assistant, built into VS Code, JetBrains and GitHub pull requests, with an enterprise admin layer.

Freemium

Cursor

Best for: AI-native editor with multi-file agent edits

A VS Code-based editor that indexes your repo so its agent can plan and apply changes across many files at once.

Freemium

Claude Code

Best for: Terminal agent for multi-step changes across a repo

Runs in your terminal and IDE, reads the codebase, runs tests and commands, and works through larger tasks you review.

Paid
Hands-on project

Ship one real ticket with an agent, reviewed and measured

Take a small ticket from your own backlog. Write the spec and context file, let your approved coding assistant implement it in small steps, prove the tests catch bugs, check dependencies and secrets, and open a pull request a colleague reviews.

What you’ll make: A merge-ready pull request with the spec, the agent's plan, your review notes and the time taken compared with a similar past ticket.

Check your own work

  • Every acceptance criterion in the spec is covered by a test
  • Tests fail when the code is deliberately broken, and the mutation check is recorded
  • Every new dependency was confirmed to exist and be maintained; no secrets are in the diff
  • The diff is small enough to review in one sitting
  • A human reviewer approved it and the time taken is logged for comparison

Try a ready-made prompt

Copy a better prompt for a real task, and see why it works better than the version most people type.

Further reading and viewing

We’ve chosen these articles, research papers, videos and guides from other trusted sources.

Research · METR · 15 min

Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity

A randomised trial where experienced devs were slower with AI but felt faster: essential reality check.

Research · DORA (Google Cloud) · 40 min

State of AI-assisted Software Development 2025

Survey of ~5,000 tech professionals on how AI amplifies strong and weak engineering practices.

Guide · OWASP Gen AI Security Project · 30 min

OWASP Top 10 for LLM Applications

The standard list of LLM security risks, from prompt injection to supply-chain issues.

Video · Y Combinator · 40 min

Andrej Karpathy: Software Is Changing (Again)

A clear mental model of how LLMs change programming, from a founding OpenAI researcher.

Course · Anthropic Academy · 60 min

Claude Code in Action

Free hands-on course on context, planning and review when working with a coding agent.

Questions about this course

What is the best AI coding assistant?

The popular choices work differently. GitHub Copilot sits in your editor and inside GitHub, Cursor is an editor built around multi-file AI edits, and Claude Code works as an agent in your terminal. The best one is the one your employer approves and that fits how you work. Good context and careful review matter more than the brand.

How do I review AI-generated code?

Review it as you would a new colleague's code, only more sceptically. Check the logic and edge cases, look for security issues such as unsanitised input, watch for duplicated code and performance traps, and check for licence problems. Confirm every dependency it adds really exists. Run the tests, and make sure they'd fail if the code were wrong.

Is it safe to use AI coding tools on company code?

Use only the tools and settings your employer approves, since some send code to outside servers. Keep secrets such as API keys out of prompts and context files, and scan for leaks. Limit what an agent can run without asking. Watch for prompt injection, where hidden instructions in a file or web page try to redirect the agent.

What is slopsquatting?

AI models sometimes suggest software packages that don't exist. Slopsquatting is when attackers publish malicious packages under those invented names, waiting for someone to install one on the AI's advice. The defence is simple. Check that every new dependency is real, well maintained and the one you meant, before it goes into your project. One lesson covers this and related risks.

How long are the lessons?

About a minute each: a short video, one question, then three takeaways and the sources. Lessons run from foundations to expert, and a three-question skills check helps you skip what you already know. The first 5 lessons are free, and Tau Pro gives you every course and lesson.

Get early access to this course

Tau opens soon. Join the waitlist and we’ll email you as soon as you can start this course.