Case study
How I used AI to refresh cultpodcasts.com without rewriting the platform: put the podcasts first, put the behaviour under test so agents can keep contributing safely, and ship across backend, API, and front-end several times a day.
The public site already indexed 80,000+ episodes. What needed changing was how that catalogue is presented: put the podcasts first, and use browse patterns people already know from media sites (featured episode, day rails, Discover), which may look a bit like a streaming UI, but only because those patterns work for browsing. Dual theme, docked search, and motion sit behind that, not on top of it.
Same stack underneath: Angular SSR/PWA, Cloudflare Workers and Durable Objects, Azure Functions, CosmosDB. Initial delivery took about five days to cutover; the next ~four days kept roughly five to seven known UI tasks in parallel on that base, washing out older front-end patterns so the next iteration is cheaper.
Hundreds of new C# business-rule tests (and matching TypeScript coverage on the website and API worker) are not for a coverage score. They are BDD-style executable rules so an AI agent can read how the system is supposed to behave, contribute more of the implementation, and fail fast when a functional change drifts. Indexing, enrichment, URL submission, hero promote, Spotify pagination: the rules live next to the code.
Spotify pagination is a concrete example: it was improved because of API quotas. Wrong paging burns quota; the rules lock the intended behaviour so the next agent change cannot quietly reintroduce an expensive walk.
Features do not land in one repo. A single change often touches the C# backend (indexer, enrichment, Azure Functions), the API layer (Cloudflare Workers, Auth0, OpenAPI), and the Angular front-end. With one developer orchestrating several features at once, the only workable rhythm is ship each piece when it is ready (multiple production deployments a day across those systems) rather than batching until everything is "done".
AI (Cursor agents against the live repos) makes that orchestration practical: I set intent and review; agents draft across the stack; the business-rule suite catches functional mistakes early. Old UI code gets washed away where it blocks faster front-end iteration; Auth0 and OpenAPI keep the API contract honest while the public surface moves.