Articles
Guides to routing, reviewing and paying for AI inference, from the team behind Inference Bureau.
Auditor
From the ticket to the merge: everything Auditor does now
Auditor now plans and builds from Linear and GitHub issues, answers questions, runs merge checks, writes tests, resolves conflicts and reports what it did.
AuditorStandard vs Thorough: what the extra pass finds, and what it costs
Auditor's two review depths side by side: what each one reads, which seeded bugs only Thorough caught, and what it cost per case in our eval.
AuditorHow we measure an AI code reviewer: 22 seeded pull requests, a scoring rule and every number
The eval behind Auditor's deep review: 22 pull requests with bugs we planted, how a finding counts as a hit, and every result from three runs.
AuditorReviewing theme pull requests for twenty client stores: an agency setup
How an agency sets up Auditor across many client theme repositories: one store each, shared rulesets, previews for sign-off, caps and client access.
AuditorOne instruction, every repository: Auditor automations
Automations run Auditor's agent on a schedule, a pull request, a push, a failed check or an issue, once per repository in scope, within limits you set.
AuditorRenaming a Shopify setting id deletes what merchants set: what each schema change loses
Each kind of Shopify theme schema change, what it does to values merchants saved on the store, and the safe way to make it.
AuditorWhat CI/CD for Shopify should look like, and what we've built so far
On Shopify, half of what a change can break lives on the store, not in git. What CI/CD for themes and apps should check, and what we've built so far.
AuditorCatch the merchant content a theme change would break, before it merges
Removing a section, block, setting or template can silently wipe what merchants built. Auditor checks the live store and says what would be lost.
AuditorPreview every Shopify theme pull request on a real store
Auditor deploys each theme pull request to an unpublished theme on your store, so reviewers click through the change instead of reading a Liquid diff.
Host apps and functions next to your model traffic. Routing picks each turn's model by your quality floor, and Auditor reviews every pull request against your own rules.