The archive · AI & Models · Technical decision · 2025
nokode: an LLM ran a CRUD web app with zero code — 300–6000x too slow
A weekend experiment gave an LLM three tools and zero app code; the CRUD app worked, but requests were 300–6000x slower and 100–1000x costlier.
nokode
What the business is
nokode is a weekend proof-of-concept by solo developer Sam Rolken: an HTTP server with no routes, controllers or business logic that sends each request to an LLM armed with three tools — a SQLite database, a web response generator and a markdown feedback memory — and lets the model decide the schema, UI and API from the URL path alone.
How it started
Sam Rolken built it in a few hours one weekend in late October 2025, after noticing the industry pointed at AI writing code — Claude Code, Cursor, Copilot — while the bigger question was whether code needs to be written at all. He chose a contact manager as the target because most software is CRUD dressed up differently, and expected the experiment to fail spectacularly.
What happened
The LLM did everything asked of it: it designed sensible SQLite schemas with proper types and indexes, returned HTML or JSON from the path alone, persisted data across restarts, produced injection-safe parameterized SQL and implemented user feedback like 'make buttons bigger'. It was also catastrophically slow — 30–60 seconds per request against 10–100 ms for a traditional app — cost $0.01–0.05 in API tokens per request, forgot what UI it generated seconds earlier and turned hallucinated SQL into 500 errors; even the prompt trick '⚡ THINK QUICKLY' made it slower. Rolken posted the results as a Show HN on 2025-11-01, drawing 436 points and 324 comments split between 'nondeterministic apps are dangerous' and 'software already changes under users'.
How it ended up
The experiment ended with a verdict rather than a product: the capability exists and the obstacles — speed, cost, consistency, reliability — are problems of degree, not kind, as inference improves roughly 10x per year and costs head toward zero. No company or product followed in the material; the repo remains live as the artifact of the test.
Background
nokode is a weekend experiment by solo developer Sam Rolken that asks what happens if you skip code generation entirely: an HTTP server with no routes, controllers or business logic sends every request to an LLM equipped with three tools — run SQL on SQLite, return any HTTP response, and save user feedback to a markdown memory. The AI infers the schema, page and API from the URL path alone.
Rolken expected the test to fail spectacularly, and in one sense it did: every click took 30–60 seconds against 10–100 ms for a traditional app (300–6000x slower), each request burned $0.01–0.05 in API tokens (100–1000x costlier), the model forgot its own UI within seconds and hallucinated SQL into 500 errors. In another sense it worked: forms submitted, data persisted across restarts, APIs returned valid JSON, and the model invented sensible schemas, safe parameterized queries and responsive layouts with no examples.
His conclusion framed the result as a timing problem rather than a capability problem: inference is improving roughly 10x per year, costs are heading toward zero, context is growing and errors are dropping, so 'AI just does the thing' may be closer than 'AI helps write code.' The Show HN on 2025-11-01 drew 436 points and 324 comments, and the repo — still live with 416 stars at the 2026-09-04 crawl — remains the artifact of the question.
What has to be true
- Rolken tested the strongest version of the idea — no application code at all — so a pass meant the capability was real rather than an artifact of clever scaffolding.
- The experiment produced measured numbers instead of opinions: 30–60 second latency, $0.01–0.05 per request and seconds of UI memory made the gap concrete.
- Honest framing ('It works. That's annoying.') earned a 324-comment debate about determinism, trust and whether post-code software is even desirable.
- Treating the gaps as problems of degree — inference ~10x per year, cost toward zero — gave the conclusion a shelf life instead of a timestamp.
What can be applied
Zero-code apps already work; what blocks them is unit economics — 300–6000x slower and 100–1000x costlier requests — so the future arrives only as fast as inference price-performance improves.
Aftermath
As of 2026-09-05 the nokode repository is still live and MIT-licensed, carrying 416 stars and 38 forks from its single commit, with no further development, company or product visible in the material — the experiment stands as the artifact of the question it asked.
Sources
- Show HN: Why write code if the LLM can just do the thing? (web app experiment)
- samrolken/nokode — a web server with no application logic
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card