How to Prepare Your Codebase for an Agentic Coding Workshop

By jocapps4 min read

Short answer: A codebase is ready for an agentic coding workshop when a fresh clone builds and tests with one command, a short context file explains structure and conventions, ten real tasks are prepared, and access, licenses and data rules are confirmed. Fixing these points two weeks ahead turns the workshop day into real work instead of setup troubleshooting.

On this page

Why preparation decides the workshop

Coding agents learn about a project the same way a new hire does: by reading files and running commands. If a new engineer needs three days and two colleagues to get the project running, an agent will struggle too, and so will a workshop. Hours spent on broken builds and missing credentials are hours not spent learning to delegate.

The good news: everything that makes a codebase agent-ready also helps human onboarding. The preparation pays off long after the workshop.

Two weeks before: the build and test baseline

Start with the single most important property: a fresh clone must build and run tests with one documented command, without tribal knowledge. Test this on a clean machine or container, not on a senior engineer's laptop that has years of accumulated configuration.

  • One command to install dependencies, for example a make target or package script
  • One command to run the unit test suite, ideally finishing in under five minutes
  • A linter and formatter configured in the repository, runnable from the command line
  • Example environment file with dummy values for every required variable
  • Local or containerized database with seed data, no dependency on shared staging
  • Pinned language and tool versions so agent and human run the same toolchain

One week before: context and tasks

Write a first context file, called CLAUDE.md for Claude Code or AGENTS.md for Codex and several other tools. Keep it to one or two screens: purpose of the system, folder map, build and test commands, conventions, and areas to avoid such as generated code or vendored libraries. The workshop will refine it, but a draft saves an hour of the session.

Then select tasks. Ten real tasks from the backlog give participants something meaningful to work on. Mix sizes and types deliberately.

  1. Three small bug fixes with a clear reproduction, ideally with a failing test possible.
  2. Two test-writing tasks for modules with low coverage.
  3. Two small features that touch two or three layers, such as API plus UI.
  4. One mechanical refactor, such as replacing a deprecated function across files.
  5. One documentation task for a confusing module.
  6. One stretch task that is deliberately ambiguous, to practice clarifying requirements.

Need a hand? neooutput helps with exactly this. Book a call for free and get a clear plan in 15 to 30 minutes.

Access, licenses and data rules

Confirm that every participant has an active seat or API access for the chosen agent, and test login from the network used on the workshop day. Corporate proxies and firewall rules block model providers surprisingly often. If participants work on locked-down laptops, check that they can install CLIs or IDE extensions, or arrange prepared development containers.

Clarify the data policy in writing: which repositories may be used with which provider, whether a zero-retention or no-training plan is required, and how secrets are excluded. Remove real credentials from the repository and from any files the agent might read. Use test API keys and sandbox accounts for third-party services.

Day before: the smoke test

Run a final smoke test with one person who represents a typical participant. Clone the repository fresh, start the agent, let it read the context file and complete one small prepared task, including running the tests. If this takes more than 30 minutes, fix the bottleneck before the workshop starts.

Also create a workshop branch strategy. Each participant or pair works on their own branch from a known commit, and nothing merges to the main branch without review. Agree with the trainer how pull requests from the session will be reviewed and merged afterward, so good work does not die in stale branches.

Get help with preparation

neooutput includes preparation in its agentic coding workshops: a scoping call, a review of your repository's agent readiness, help with the first context file and selection of lab tasks from your backlog. The AI Code Engineers who run the sessions flag blockers early so the workshop day is used for learning, remote or on-site and in English or German.

If the readiness review shows that an AI-generated codebase needs a deeper audit or rescue before the team can work on it productively, apprat.io offers code audits and Prototype-to-Production work with human review of every change. A free 15-minute call with neooutput is the easiest way to check how ready your codebase already is.

Frequently asked questions

What if our codebase has no tests?

You can still run a workshop, but start by having participants use the agent to write characterization tests for one critical module. This is an excellent exercise in itself and creates the safety net later tasks need. Without any tests, agents cannot verify their changes, so expect more manual review.

How long should the context file be?

Short. One or two screens is enough for most projects. Focus on commands, structure, conventions and known traps. Long files cost context on every request and the agent may ignore buried instructions. Monorepos can use additional context files in subfolders for module-specific rules.

Can we use a demo project instead of our real code?

A demo project is fine when legal approval for your code is pending, but the learning transfer is weaker. Participants miss the realistic challenges of large files, legacy patterns and unclear requirements. If possible, use a less sensitive internal repository rather than a toy project.

Do participants need admin rights on their laptops?

Not necessarily. They need to install a CLI or IDE extension and run the project's build. If admin rights are restricted, prepare development containers or cloud workspaces in advance, and verify that the agent can reach the model provider from that environment.

Also useful for this

Talk to an AI Code Engineer, free of charge

Tell us where you are and where you want to go. In a short call we map out the fastest path, whether that is a workshop, a launch in the App Store and Google Play or a full AI-native product.

No obligation. English or German. Run by jocapps, ISO 27001 and ISO 9001 certified. Prefer the phone? 24/7 call support: +49 761 519 751 50

Book a call for free

Related guides

Book a call for free