Field notes from making a living without writing a line of code (ag404labs.com)

A former career developer describes how he now makes a living by using Claude as a live co‑developer: copying error messages into the model, pasting fixes back into production, and billing clients based on value delivered. In a week he patched a Django loan‑approval bug caused by a math typo, generated a HIPAA‑safe audit‑logging middleware (initially too verbose, then corrected after prompting), resolved a reconciliation bug caused by centavos/pesos unit confusion, translated Portuguese bank docs to restore payments, and rewrote slow SQL (fixing n+1 queries, adding indexes and eliminating a Cartesian join) to cut page load from 8s to 0.3s. He invoices $60–80k/month while running on a modest cloud + LLM subscription stack, emphasizing speed over long‑term architecture. The story matters because it illustrates a practical trajectory for LLMs in operations: they’re already good enough to triage, generate and iterate production fixes rapidly, and handle cross‑lingual/API cruft—redefining the consultant role into an orchestrator, translator and responsibility‑taker. But it also underscores limits: LLMs accelerate delivery without resolving political, regulatory or design tradeoffs, and they institutionalize “good‑enough” code and tech debt. The model raises sustainability and ethical questions—clients may someday bypass intermediaries, yet many will still pay for human judgment, accountability and someone who knows what to ask.