AI agents love to have every part of your application at hand: the logic, the templates, the server config, and yes even your database is best kept locally, so the agent can check and update it directly.
For years the industry told us to split everything apart. Frontend here, API there, auth as a SaaS, database in the cloud, logs somewhere else, and a pipeline to glue the pieces back together. Microservices, headless CMS, serverless functions — each one another hop, another dashboard, another place where an AI coding agent has to guess what the rest of your app looks like.
Agents are not great at guessing across five repos and three paid services. They are great when the whole thing sits in one folder.
Give an agent a classic monolith — PHP (or whatever) that renders pages, talks to a local SQLite or MySQL, and lives next to its config — and suddenly it can see the application. Logic, templates, routes, schema, even the data. No stubbing out half the stack. No “check the other repository.” No waiting for a remote API that the agent cannot inspect.
That is why the boring stack is back again. One codebase. One server. One database you can open with a client. The same setup that powered the old internet turns out to be the setup AI works with best: small enough to hold in context, complete enough to change end to end.
I am not saying microservices were always wrong. I am saying that for most sites — the ones we used to ship as a handful of files on shared hosting — the monolith never stopped being the honest answer. AI just made the honesty visible again.
So keep it local when you can. Let the agent read the whole app. Ship something you still understand without a whiteboard full of boxes and arrows. The previous web already knew that trick. We are only remembering it.