Pipeline Automation

At scale, manual pipeline management — running generation, routing to critic, handling approvals, updating metadata, deploying assets — is the bottleneck. Automation handles the repetitive mechanics so that human attention focuses on creative decisions and quality judgment.

Automate these steps: queue management (priority ordering, rate limiting, retry on failure). Template instantiation (parse the CSV, fill the template variables, dispatch to the generator). Metadata creation (auto-populate generated_at, prompt, style_version from the template context). Cohort check triggering (every 10 assets, automatically dispatch a batch to the consistency reviewer). Deployment tracking (when an asset URL is referenced in code, automatically update the deployed_to metadata).

Do not automate these steps: quality approval (the critic scores, but a threshold pass should still be reviewed for cohort consistency). Style specification updates (style evolution requires creative judgment). Template design (the locked and variable elements require design thinking). The automation/judgment boundary is clear: automate mechanics, preserve human (or senior agent) judgment for creative and quality decisions.