Simon Willison originally believed that coding agents could only produce things that looked like games but weren't actually fun. He intentionally tested this assumption with a vague instruction; Claude, running in Ultracode mode, conducted over 1,000 simulated matches, using the results to rebalance units and mechanics, ultimately producing what he considers a genuinely enjoyable game. This experiment reveals under what conditions vague quality instructions are effective versus when they merely result in superficial feature stacking.
In an AIE CODE podcast episode, Monzo engineers Felipe Felisola Caso and Francesco Galletta share the evolution of their in-house AI coding agent. The tool has expanded from a simple Slack interaction to authoring over 10% of all merged pull requests across the company and executing more than 1,800 tasks daily. The discussion highlights the 'multiplier' architecture enabling this scale, including an MCP gateway for internal tools, a sandboxed orchestration system compliant with banking regulations, and a platform strategy that treats the agent as a fundamental building block.
Based on a survey of 1,528 DevSecOps professionals, GitLab reports that AI coding agents have improved individual developer productivity, but active development accounts for only 10–20% of software delivery lead times at large enterprises, with the bottleneck having shifted to review, approvals, and environment waits. The company's AI-Native Engineering Playbook proposes a four-stage autonomy ladder whose core rule is that each promotion must be backed by evidence from the rung below, and warns that a nominally-in-the-loop reviewer who rubber-stamps everything is worse than no checkpoint at all.
In 'My Obituary for Programming,' Jeremy Scheff draws a specific distinction: if what you enjoy is solving logic puzzles themselves, coding agents bring a real loss; if what you enjoy is getting systems built, large models are a continuing gift.
Equals Chief Product Officer James Simcox stated at Okta's annual conference that over 92% of the company's code is now written by AI. He detailed the shift from broad AI adoption to selective use, emphasizing narrower permissions for agents compared to humans and the need to log agent reasoning to mitigate large-scale error risks.