What to set up so your coding agent can work without you watching every step.
You use a coding agent every day, and you still spend most of that time watching it: restating what you meant, fixing the same mistakes, running the app to see whether it works, and reading every diff.
This four-hour live session covers how experienced users work with coding agents in late 2026, organized by what you can stop doing yourself and what has to be true before you can. I work through my own setup on real repositories (Claude Code and Codex, skills, project instructions, verification scripts, parallel sessions), show where each piece came from, and where it still fails.
The last part of the session is yours: bring a task where your agent let you down, and we work out which piece was missing.
What you’ll learn
Hand more of your work to the agent, and know what has to be in place before you do.
Have the agent question you before it starts, and write down what done means as checks.
Hand over a spec whose acceptance criteria a script can check.
Move what you keep correcting into lint rules, types and conventions the agent runs into on its own.
Replace the deterministic steps of a task with scripts, so the agent does not improvise them.
Give the agent a way to run the code and inspect the result, beyond the test suite.
See tests pass on wrong output, and check that a verifier rejects a known-bad result.
Mine your past sessions for the corrections you make most often.
Choose between a skill, a script and a rule, and drop skills that do not help.
Run parallel and background sessions, and start routine work from issues or chat.
Review a sample of the changes, fix the setup rather than the single diff, and know when that is unsafe.
Workshop agenda
Where coding agents are now
Terminal, desktop and cloud agents, what changed since the summer, and what you can stop doing that was needed six months ago.
Getting your intent across
Letting the agent question you before it starts, a spec with checkable acceptance, and why the exact words matter.
Constraints the agent runs into
Lint rules, types and conventions in place of more instructions, and scripts for the parts that never need judgment.
Verification
Letting the agent run and inspect its own work, tests that pass on wrong output, and how to check the checks.
Skills from your own corrections
Mining past sessions for what you keep correcting, choosing between a skill, a script and a rule, and dropping skills that do not help.
More than one agent
Parallel and background sessions, routines and triggers, reviewing by sampling, and when that is not safe.
Your failures
Bring a task where your agent let you down; we work out which piece was missing.
Learn directly from Stefan
Stefan Jansen
Uses coding agents daily across software, research, and publication workflows.
Who this workshop is for
You use Claude Code, Codex or Cursor on real work daily, and still spend most of that time supervising it.
What's included
Lifetime access
Go back to course content and recordings whenever you need to.
Community of peers
Stay accountable and share insights with like-minded professionals.
Certificate of completion
Share your new skills with your employer or on LinkedIn.
Maven Guarantee
Your purchase is backed by the Maven Guarantee.
Frequently asked questions
Maven for Teams
Reimbursement
Get your company to pay
Everything L&D needs: email template, receipts, and certificate of completion.
Get reimbursedPrivate cohort
Run a cohort for your org
A dedicated cohort with a custom schedule and curriculum, tailored to your team.
Book a private cohort