Claude Code Workshop for Digital Agencies

Diese Seite auf Deutsch

Short answer

A Claude Code workshop for agencies has to go beyond prompting basics. It should cover onboarding the agent to many client codebases with instruction files, respecting client confidentiality, quoting projects when parts of the work get faster, and keeping review, testing and handover at a standard clients can rely on long after the project ends.

Why agencies need a different workshop

Most coding agent training assumes one product team working in one codebase. An agency looks different: developers switch between many client repositories in a week, each with its own stack, conventions, hosting and contractual rules. The agent has to be productive in a repository it saw for the first time this morning, and the agency has to answer to a client for every line it ships. A useful workshop is built around exactly that.

Many codebases: getting the agent productive per client

  • A CLAUDE.md per repository: the stack, the exact commands to install, build, test and lint, naming conventions and the areas the agent must not touch. Ten minutes of writing saves hours of wrong guesses.
  • Agency defaults in one place: a user-level CLAUDE.md can carry your house standards (commit style, accessibility basics, review checklist), while the project file holds what is specific to the client.
  • Reusable commands: recurring tasks like scaffolding a component, writing a migration or running an accessibility check become custom slash commands, so juniors get the same quality as seniors.
  • Permissions: deny rules in the settings keep the agent away from .env files, production deploy scripts and client infrastructure.
  • MCP servers: connect issue trackers or design tools only where the client has agreed.

Client confidentiality comes first

Before anyone runs an agent on client code, check what the contract allows. Some clients prohibit sending source code to external AI services, others require that subprocessors are listed. If personal data is involved, data protection agreements may need updating. Record per client which tools are approved, use business plans with clear data use terms and keep secrets out of prompts and repositories. This is general guidance, not legal advice – your contracts decide.

Quoting when delivery gets faster

Agents speed up some work a lot and other work barely at all. Scaffolding, boilerplate, test writing and refactoring get faster. Discovery, design decisions, stakeholder alignment and careful review do not. That has consequences for how you sell:

  1. Measure first: track estimate versus actual on agent-assisted tickets for a few projects before changing prices.
  2. Price the outcome: fixed-scope offers capture the efficiency gain; pure time-and-materials contracts pass it entirely to the client.
  3. Make quality visible: list tests, documentation and review as explicit deliverables, so faster delivery does not read as less work.
  4. Be transparent: tell clients you use coding agents and how you control them. It builds trust and avoids awkward questions later.

Quality gates and handover

Every agent-generated change needs a developer who reviews and understands it. Back that up with CI that runs tests and linters on every pull request. Claude Code can assist reviews, for example in a GitHub Actions workflow, but a human approves. At handover, the client receives code that their own team can maintain with or without agents: a current README, the CLAUDE.md as living documentation, decision records and no dependency on prompts or tools that only exist inside the agency.

How jocapps helps

Our training brand neooutput runs hands-on workshops, team trainings and 1:1 consulting on agentic coding – build-as-you-learn, so every session ends with working results. For agencies that means practicing on realistic project setups: writing instruction files, configuring permissions, running reviews and discussing how to quote. Workshops are available in English and German, on-site or remote. Our guide on Claude Code best practices is a good primer, and the agentic coding workshops page describes the formats.

Want to discuss a workshop for your agency? Book a free 30-minute intro call via the meeting planner.

Frequently asked questions

Can we use Claude Code on client projects at all?

Often yes, but check each contract first. Some clients exclude external AI tools or require approval of subprocessors, so document per client what is allowed.

Should the CLAUDE.md be handed over to the client?

Usually yes. It documents build commands and conventions, which helps the client's team whether or not they use coding agents themselves.

Will agentic coding reduce our revenue on hourly contracts?

It can, because time-and-materials passes efficiency gains to the client. Many agencies move toward fixed-scope or outcome-based offers once they have measured real time savings.

Guides on this topic

Talk to us about your project

Honest advice on your idea, prototype or app.

  • 30 minutes
  • Free
  • No obligation

Book a free call [email protected]