Workshops: Agentic Engineering Practices

Coding Agents in Large Codebases & Monorepos: Workshop

Coding agents that shine on a small project often struggle in a codebase with millions of lines, dozens of teams and a long history: they search in the wrong places, fill their context with noise and lose track halfway. This workshop shows your team how to scope context, use search and indexing deliberately and slice work so agents stay effective at enterprise scale.

Who is this workshop for?

  • Developers in large monorepos or sprawling multi-repo landscapes
  • Staff and principal engineers who shape how agents are used across the codebase
  • Developer experience and build teams working with Bazel, Gradle, Nx or Maven multi-module setups
  • Architects of long-lived enterprise systems with many modules
  • Teams whose first agent experiments in the main repository were disappointing

What you will take away

  • Understand how coding agents gather context and why large repositories overwhelm them.
  • Structure layered instruction files and module documentation so agents find the right entry point.
  • Choose between text search, code intelligence and semantic indexing for a given question.
  • Slice large tasks into context-sized units with plan files and progress notes.
  • Delegate exploration to subagents or separate sessions so the main session stays focused.

Agenda

  1. Module 1

    How agents see a codebase

    • Context windows, tool calls and why more context is not automatically better
    • Agentic search vs. pre-built indexes: Claude Code, Cursor, GitHub Copilot and OpenAI Codex compared
    • Signs of context overflow: drift, forgotten constraints, repeated work
  2. Module 2

    Making the repository navigable

    • Layered CLAUDE.md or AGENTS.md files per directory or package
    • Ownership files, module maps and architecture overviews as entry points
    • Exercise: a navigation layer for one area of your repository
  3. Module 3

    Search and retrieval

    • ripgrep, structured search, language servers and code intelligence via MCP
    • Indexing and semantic search: strengths and blind spots
    • Excluding generated code, vendored libraries and build output
  4. Module 4

    Slicing big changes

    • Breaking a cross-cutting change into context-sized steps
    • Plan files and progress notes that survive a fresh session
    • Exploring in a subagent, implementing in a clean main session
  5. Module 5

    Build and test at scale

    • Running only affected tests and targets
    • Keeping an eye on runtime and cost of long agent sessions

Why large codebases are different

In a small repository an agent can read most of what matters. In a large one it has to choose, and every file it reads competes for limited context. Without guidance it opens the first plausible match, misses the internal library everyone is supposed to use and re-implements something that already exists three directories away.

The fix is rarely a bigger model or a longer context window. It is a codebase that explains itself at the right level, a search strategy and tasks cut to a size the agent can keep fully in view.

What changes in your repository

Typical results we implement together during the day:

  • A short root instruction file pointing to deeper, package-level guidance
  • Exclusion rules for generated, vendored and binary content
  • Module maps that state which libraries and patterns are canonical
  • A template for multi-session tasks with plan, status and open questions

Confidential code at scale

Large codebases often hold a company's most valuable intellectual property. We clarify which indexing features process code on the provider's servers, how enterprise plans differ on data retention and when a sandbox mirror is the better choice. jocapps, the company behind neooutput, is ISO 27001 certified, and our AI Code Engineers work within your security policies.

Frequently asked questions

Does a bigger context window solve the problem?

Only partly. The agent still has to find the right files, and in very long contexts instructions are more easily overlooked. Good structure and well-sliced tasks pay off with every model.

Should we split our monorepo for the agents?

Usually not. Monorepos work very well with agents once instruction files, ownership and build tooling are in place. We help you decide whether specific areas need clearer boundaries.

Which tool handles large codebases best?

Claude Code, Cursor, GitHub Copilot and OpenAI Codex approach search and indexing differently, each with trade-offs. We test them on a realistic task in your repository so you can judge for yourselves.

Workshop

Get in touch

Where do coding agents get lost in your codebase today?

Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.

Prefer to talk first?

15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.

Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)

Request a training / workshop

Workshop: Coding Agents in Large Codebases & Monorepos: Workshop · 1 day

Dates that work for you (optional)

Suggestions about 2 weeks ahead – pick as many as you like

Reply usually within 24 hours · jocapps is ISO 27001 certified · Privacy