Coding with agents

Shipping software changes with a coding agent. It starts with your first session in a small repository. From there it follows one change from plan to verified diff, and it ends with working alongside a team.

0%

1. getting started
2. halfway
3. almost there
4. complete

tutorialYour first session with a coding agenttutorialKeeping API keys out of the agent's reachexplanationDeciding whether to build it at alltutorialTelling the agent where the task is and where it stopstutorialWriting project instructions the agent reads every sessiontutorialReviewing a diff you did not write, against the spectutorialReproducing a fault before the agent fixes ittutorialTests that prove behavior and docs that stay currenttutorialReading agent code as if the input were hostiletutorialSandboxing a coding agenttutorialAttacking your own agent before someone else doestutorialRunning several agent sessions in worktrees

tutorialLet the agent interview yoututorialSplitting the work into components the agent can build one at a timetutorialPlan first, then drive the change from a failing testtutorialWorking from a spec in small incrementsexplanationWhat each phase starts withtutorialA review from a fresh sessiontutorialGiving the agent a check it can run, with a limitexplanationReviewing what the agent pulled inexplanationSaying what the agent did, in the commit and the review

tutorialA prototype you throw awaytutorialKeeping every agent change easy to undoexplanationStaying the engineer who understands the systemtutorialEnforcing the team's rules with hooks and CI

tutorialA backlog and standards the agent can readexplanationWriting down how the team uses agents

Review: nothing due yet

About this course

Goals

• Ships a change with a coding agent through plan, implement and verify
• Specifies work well enough for an agent to implement and for anyone to verify
• Verifies agent-written code before trusting it
• Works with agents alongside a team

Counts

27 lessons, 106 checkpoints

Builds on

• Grounding and memory
• Delegating to an agent
• Agent risk
• Decomposing work
• Verifying outputs

Shipping software changes with a coding agent. Progress is kept in this browser; see your progress.

Lesson plan (27 lessons, 27 live)

#	Lesson	Status	Mode	Minutes	Covers	Serves	Exercise	After	Issue
Running a coding agent
1	Your first session with a coding agentcoding-with-agents/first-session	live	tutorial	35 estimate	first-session	runs-a-sessiongives-the-right-context	do
2	Keeping API keys out of the agent's reachcoding-with-agents/keys-and-secrets	live	tutorial	20 estimate	first-session	runs-a-session	do	Your first session with a coding agent
Deciding and specifying
3	Deciding whether to build it at allcoding-with-agents/deciding-what-to-build	live	explanation	25 estimate	specification	judges-worth-buildingdefines-success	judge	Your first session with a coding agent
4	Let the agent interview youcoding-with-agents/agent-interviews-you	live	tutorial	35 estimate	specification	is-interviewed-before-planning	do	Deciding whether to build it at all	#677
5	Splitting the work into components the agent can build one at a timecoding-with-agents/decomposing-the-work	live	tutorial	25 estimate	specification	decomposes-into-componentsdesigns-the-check	do	Deciding whether to build it at all	#150
6	A prototype you throw awaycoding-with-agents/prototype-to-throw-away	live	tutorial	30 estimate	specification	settles-with-a-prototype	do	Splitting the work into components the agent can build one at a time, Let the agent interview you	#679
Plan, implement, verify
7	Plan first, then drive the change from a failing testcoding-with-agents/plan-then-test	live	tutorial	30 estimate	workflow	runs-a-session	do	Your first session with a coding agent, Splitting the work into components the agent can build one at a time
8	Working from a spec in small incrementscoding-with-agents/spec-driven-increments	live	tutorial	30 estimate	workflow	works-in-increments	do	Plan first, then drive the change from a failing test	#155
9	Keeping every agent change easy to undocoding-with-agents/reversible-changes	live	tutorial	35 estimate	workflow	keeps-change-reversible	do	Working from a spec in small increments	#157
Context engineering for code
10	Telling the agent where the task is and where it stopscoding-with-agents/scoping-and-referencing	live	tutorial	25 estimate	context	gives-the-right-context	do	Plan first, then drive the change from a failing test	#160
11	Writing project instructions the agent reads every sessioncoding-with-agents/project-instructions	live	tutorial	35 estimate	context	gives-the-right-contextkeeps-understanding	do	Telling the agent where the task is and where it stops	#163
12	What each phase starts withcoding-with-agents/what-a-phase-starts-with	live	explanation	20 estimate	context	chooses-what-a-phase-starts-with	judge	Writing project instructions the agent reads every session	#681
13	Staying the engineer who understands the systemcoding-with-agents/keeping-understanding	live	explanation	25 estimate	workflow	keeps-understanding	judge	Writing project instructions the agent reads every session, Keeping every agent change easy to undo	#166
Verifying agent work
14	Reviewing a diff you did not write, against the speccoding-with-agents/reviewing-the-diff	live	tutorial	30 estimate	verification	reviews-others-code	judge	Keeping every agent change easy to undo	#169
15	A review from a fresh sessioncoding-with-agents/review-in-a-fresh-session	live	tutorial	25 estimate	verification	uses-a-fresh-review	do	Reviewing a diff you did not write, against the spec	#680
16	Reproducing a fault before the agent fixes itcoding-with-agents/observing-and-debugging	live	tutorial	25 estimate	verification	observes-and-debugs	do	Reviewing a diff you did not write, against the spec	#172
17	Giving the agent a check it can run, with a limitcoding-with-agents/self-checking-loops	live	tutorial	25 estimate	verification	automates-the-checkprefers-gates-to-opinions	do	Reproducing a fault before the agent fixes it	#174
Quality with agents
18	Tests that prove behavior and docs that stay currentcoding-with-agents/tests-and-docs-that-hold	live	tutorial	30 estimate	quality	reviews-others-codeautomates-the-check	judge	Giving the agent a check it can run, with a limit	#176
19	Reviewing what the agent pulled incoding-with-agents/dependency-review	live	explanation	30 estimate	quality	screens-for-security	judge
20	Reading agent code as if the input were hostilecoding-with-agents/security-review	live	tutorial	30 estimate	quality	screens-for-security	do	Tests that prove behavior and docs that stay current	#178
Sandboxing and red-teaming
21	Sandboxing a coding agentcoding-with-agents/sandboxing	live	tutorial	25 estimate	first-session	runs-a-session	do	Keeping API keys out of the agent's reach	#18
22	Attacking your own agent before someone else doescoding-with-agents/red-teaming-your-agent	live	tutorial	30 estimate	verification	screens-for-security	judge	Sandboxing a coding agent, Giving the agent a check it can run, with a limit	#180
Agents in a team
23	Saying what the agent did, in the commit and the reviewcoding-with-agents/attribution-and-review-norms	live	explanation	25 estimate	team	attributes-honestlyfollows-team-norms	judge	Keeping every agent change easy to undo, Reviewing a diff you did not write, against the spec, Reading agent code as if the input were hostile	#183
24	Enforcing the team's rules with hooks and CIcoding-with-agents/ci-and-hooks	live	tutorial	30 estimate	team	follows-team-normssets-team-practice	do	Giving the agent a check it can run, with a limit	#186
25	A backlog and standards the agent can readcoding-with-agents/backlog-and-standards	live	tutorial	40 estimate	team	keeps-the-backlog-reachableturns-findings-into-rules	do	Enforcing the team's rules with hooks and CI	#678
26	Running several agent sessions in worktreescoding-with-agents/parallel-sessions	live	tutorial	30 estimate	team	runs-parallel-work	do	Enforcing the team's rules with hooks and CI	#190
27	Writing down how the team uses agentscoding-with-agents/team-practice	live	explanation	25 estimate	team	sets-team-practice	judge	Running several agent sessions in worktrees, Saying what the agent did, in the commit and the review	#193