Lesson
Short theory and a narrow lab objective.
video · 6 minPractice labs for agentic engineering
Practice labs for AI-assisted engineering.
Learn patterns. Run agents. Inspect evidence. Compare methodology tradeoffs.
Prepared sandbox labs produce logs, diffs, and rubric evidence so engineers can judge the workflow, not just the tool.
2 free labs in CatalogManaged credits shown in pricingNo card to start
The loop
Lesson → Sandbox → Agent run → Evaluation → Comparison
Sandbox
A frozen checkpoint materializes.
audit-log@v1.4Agent run
Guided actions produce logs and diffs.
agent traceEvaluation
Rubrics and evidence decide progress.
score + gatesComparison
Methodology tradeoffs stay tied to the same lab.
3 packsA taste of a real lab
Lab 1 · Run the first verify gate
free preview · runs through the guided sandbox path
Goal
Practice lab in Free Orientation: What the portal is, how labs work, and why verification comes before confidence.
35mestimated time
introdifficulty
completedcatalog state
Run timeline
- Sandbox materializedcheckpoint audit-log@v1.4
- Agent run capturedlogs, diffs, and commands
- Tests summarizedgreen path plus warnings
- Evaluation publishedrubric gates and next step
12 patterns
Patterns you'll practice
Verification-first feedback loop
Lesson in Free Orientation: What the portal is, how labs work, and why verification comes before confidence.
Run the first verify gate
Practice lab in Free Orientation: What the portal is, how labs work, and why verification comes before confidence.
Layered steering
Lesson in Foundation & Steering: Project maps, local instructions, prompt corridors, and durable operating loops.
Local instructions as invariants
Lesson in Foundation & Steering: Project maps, local instructions, prompt corridors, and durable operating loops.
Audit a repo instruction stack
Practice lab in Foundation & Steering: Project maps, local instructions, prompt corridors, and durable operating loops.
The compact spec ladder
Lesson in Spec-Anchored Delivery: Turn rough ideas into scoped implementation slices and evidence-backed plans.
3 methodology packs
Methodologies, judged honestly
Spec-anchored delivery
Ground the run in the prepared spec and current slice boundary.
Verification-first
Prioritize executable evidence before any broader workflow expansion.
Context audit
Inspect instructions, bounded context, and dependency readiness first.
Honest tradeoffs
The portal does not declare a universal winner. Every comparison ends with when to use a workflow and when not to.
What this is not
This is not an IDE.
The learner manages agent runs, not browser files. They read generated diffs, logs, test output, and evaluation evidence before deciding whether the methodology earned its weight.
No browser code editor
No save-file UX
No public leaderboard
Read diffs and logs
read-only · src/audit-writer.ts
2 changes
@@ generated by agent @@
- write(event) {
+ async write(event: AuditEvent) {
+ assertActor(event.actor);
return db.insert(event);
read-only. that's the point.
For engineering leaders
Standardize AI-assisted engineering across your team.
Repeatable labs, a shared rubric, and evidence that engineers learn the patterns. Not just licenses for a tool, but a way of working you can roll out.
