# Agentic Code Review pitch

> A first-pass deck for a stateful, model-agnostic code review service

- HTML version: https://robbiepalmer.me/projects/agentic-code-review/deck
- Project: Agentic Code Review (https://robbiepalmer.me/projects/agentic-code-review.md)
- Slides: 8

## Slide 1

# Code arrives faster than reviewers can read it

Agentic Code Review gives every pull request a dependable second opinion.

## Slide 2

## The bottleneck moved

Coding agents can produce parallel changes. Human attention did not scale with them.

More generated code makes fast, trustworthy review infrastructure more valuable.

## Slide 3

## Managed reviewers keep changing the deal

* Quotas can stop reviews halfway through a busy week.
* Every push can pay again for context the pull request already supplied.
* Useful repository learning becomes difficult to move to another provider.

## Slide 4

Own the review loop. Rent the models.

### Keep

Pull request state, repository policy, accepted findings, scorecards, and evaluation data.

### Replace freely

Models, inference providers, and review depth for each change.

## Slide 5

## One pull request, one evolving review

```mermaid
flowchart LR
A[Pull request] --> B[Risk policy]
B --> C[Reviewers]
C --> D[Reconcile]
D --> E[GitHub review]
E --> F[Disposition data]
```

## Slide 6

## A stable finding contract keeps models interchangeable

```ts {3-5}
type Finding = {
  path: string
  line: number
  severity: "blocking" | "material" | "minor"
  evidence: string
  proposedFix?: string
}
```

The highlighted lines are the shared policy boundary for every model adapter.

## Slide 7

## The first product stays deliberately narrow

It starts with this site's pull requests, records what happened, and learns which findings developers accepted.

The longer product brief and [architecture decisions](/projects/agentic-code-review?tab=adrs) remain available outside the deck.

## Slide 8

## Success looks boring

* Every material pull request gets reviewed.
* Repeat pushes do not buy the same context twice.
* A better model joins through an adapter, not a migration project.

---

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