Direct the Agent
Plan a feature, give the agent context, test it and verify the result
Session three moves beyond one-shot code generation. You define the outcome and evidence, the agent reads the repository and proposes a plan, then it implements, runs checks and proposes fixes while you review the result and hold the product decisions.
Dates TBD — evening workshops followed by a longer capstone hackathon at Workuity Biltmore. Join Discord or follow the Meetup calendar for the next cohort.
Or join the waitlist — the dates come to you.
Venue sponsor: Workuity. Format: the host builds live on screen. Bring a laptop and work along, or just watch — nobody is put on the spot.
Why this session
Agentic engineering works when the agent can see the system, use tools and recognize done. Tests, logs and explicit acceptance criteria turn the loop from 'looks right' into evidence you can inspect.
What we cover
- 01
The agentic loop
Plan, execute, observe and iterate — what the agent owns, what the human owns and when to interrupt.
- 02
Give the repository memory
Write project instructions, reusable skills, conventions and commands once so future sessions start with the right context.
- 03
Scope a feature
Acceptance criteria, edge cases and a short implementation plan that can survive contact with the existing codebase.
- 04
Tests as the specification
Types, lint, build checks, unit tests and browser tests; make one check fail, read the output, fix it and rerun the gauntlet.
- 05
Verify the outcome, not just the code
Use browser and integration tools to complete the user journey, visually inspect the interface and query the resulting database row and defaults.
- 06
Let failure steer the work
Read terminal output and server logs, have the agent diagnose the evidence and propose a fix, then verify the result instead of guessing.
- 07
Review before you ship
Inspect the diff for security, unintended scope and user-facing regressions, then leave a clean checkpoint.
You leave with
A tested feature delivered through a reusable plan-execute-observe-iterate workflow, plus project instructions your next agent session can trust.
To follow along
A laptop and your deployed project from session two. Bring access to its repository, database and deployment logs so the agent can work across the whole loop.
Afterward
Nothing is required. It compounds if you run the same work on your own product. The suggested output:
03-agentic/ — feature brief, agent plan, tests, verification log and the final reviewed diff.
Everything is open source. Missed the session? It's archived in Discord.
The session kit
Slides, a follow-along guide and skill files. Free to download, room or no room.
The community online
Recaps, archive links, questions between sessions, everyone else's work in progress.