Engineering With Coding Agents

Four series of The Pragmatic Stack covered the architecture and operations of LLM-powered systems: the discipline inside the agent, the systems around it, the operations program that keeps them running in production, and the .NET runtime that implements all of the above.

All four series were about systems that contain agents, or agents that serve users.

Series 5 is about a different kind of agent: one that opens your repository, reads your codebase, and makes commits.

Coding agents are not a new developer experience. They are a new actor in the software delivery system. And that system — the repository, the task backlog, the branch strategy, the test suite, the review process, the deployment pipeline — was not designed with an automated author in mind. Some of it will need to change.

Engineering With Coding Agents is six articles on exactly what needs to change, and how.

This is the fifth series in The Pragmatic Stack:

• Architecting Agents (Series 1) covered the discipline inside an agent
• System Design, Reimagined (Series 2) covered the systems around the agent
• Shipping AI Systems (Series 3) covered how those systems are operated in production
• AI for the .NET Architect (Series 4) covered what all of the above looks like in .NET
• Engineering With Coding Agents (Series 5) covers what your delivery system needs to handle agents as authors

What you will get out of it:

• A framework for making any repository legible to a coding agent without writing a small novel of instructions
• A task contract format that prevents vague intent from becoming production code faster than the team can catch it
• An isolation model for parallel coding agent work: one branch, one worktree, one environment per task
• A six-layer verification ladder that distinguishes software correctness from LLM benchmark performance
• A code review approach designed for generated diffs at scale: risk-ordered, provenance-first, test-aware
• A seven-stage delivery loop that produces accepted, verified, reviewable changes instead of raw generated output

The articles:

1. The Repository Is the Prompt: Preparing a Codebase for Coding Agents
2. The Ticket Is the Contract: Specifying Work for Coding Agents
3. One Agent, One Worktree: Isolation and Concurrency for Coding Agents
4. Tests Are the Verification Layer: Proving Agent-Generated Changes
5. Reviewing Code Nobody Typed: Human Judgment in an Agentic Workflow
6. The Pull Request Is the Product: A Delivery System for Coding Agents

Read them in order if you are building the delivery system from scratch. Jump to the article that maps to the constraint you are currently hitting if you are already partway through adoption.

The agent writes the code. The delivery system decides what reaches production.