Strategy and Data with AI
Today I tried something that sounded impossible: building a web app with zero application code.
No routes, no controllers, no logic files... just an LLM with three simple tools: - a database it could query - a web response function to generate HTML or JSON - a small memory file for feedback
For every HTTP request, the server just asked the model, "What should I do?"
To my surprise, it worked. The AI generated a full CRUD contact manager on its own. It invented database schemas, wrote safe SQL queries, created responsive layouts, and even adapted to feedback like "make the buttons bigger" or "add a search box." Every piece of behavior emerged from the model interpreting intent, not code.
Of course, it was wildly impractical. Each click took 30–60 seconds. Costs were hundreds of times higher than a normal app. The model forgot designs between requests and hallucinated errors. It was like running a Ferrari engine in a lawnmower.
What struck me was that the failure was quantitative, not conceptual. The system proved that the logic layer of software can already be handled by AI. The gaps are speed, cost, and memory... which are all improving exponentially.
That experiment changed my perspective. We've been focused on "AI that writes code," but that might already be an outdated framing. The real shift will be when we skip code entirely, when the interface is just intent and execution.
Business takeaway: If AI can already reason through application logic, even inefficiently, the first organizations to reimagine workflows for a post-code world will hold an enormous advantage when the performance curve catches up.
hashtag#AI hashtag#FutureOfSoftware hashtag#NoCode hashtag#LLM hashtag#DigitalStrategy hashtag#Experiment