# ADR 011: Coding agents and AI review in the base layer

- HTML version: https://robbiepalmer.me/projects/personal-engineering-platform/adrs/011-agentic-development-baseline
- Project: Personal Engineering Platform (https://robbiepalmer.me/projects/personal-engineering-platform.md)
- Status: Accepted
- Date: 2026-09-15
- Initiatives: Semi-autonomous Software Development (https://robbiepalmer.me/initiatives/semi-autonomous-software-development.md)

## Context

Coding agents and automated review now participate in work across every project
that adopts the base layer. Keeping their decisions in the Personal Knowledge
Graph makes platform-wide changes look local and encourages projects to drift
between different agent and review setups.

Both roles use more than one tool at a time. Codex and Claude Code provide
independent capacity and different working styles. Pull Requests receive
separate opinions from managed reviewers and the stateful in-house reviewer.
A single-selection slot cannot describe either set honestly.

## Decision

Add coding agents and AI reviewers as required base-layer slots. Allow these
slots to hold several accepted selections concurrently.

Select Claude Code and Codex as coding agents. Select CodeRabbit, Greptile,
Codex code review, and Agentic Code Review for Pull Request review. Projects
that track the base layer receive additions, removals, and replacements as one
shared policy change.

## Consequences

Agent instructions and review policy can change once for every tracking
project. Multiple providers reduce the chance that one outage, allowance, or
blind spot stops delivery.

The baseline consumes more review capacity and produces overlapping comments.
The review loop still has to dismiss weak findings and prevent several tools
from turning one issue into repeated work. Removing one tool closes only its
selection period; it does not replace the remaining set.

---

Markdown index of this site: https://robbiepalmer.me/llms.txt