AI coding agents have broken the operating model of software interfaces. Aurelie Radom's lead piece frames the problem precisely: when an agent decides which files to change, which tests to run, and which commands to execute, the interface stops being a tool you operate and starts being an actor you supervise. The five questions she surfaces, what did the system understand, what is it doing, what has it changed, what can you stop, what needs review, are not rhetorical. They are the unsolved UX problem of the agentic era.
Three pieces in this edition push that problem into concrete territory. Dan Maccarone argues AI has stripped engineering of its gatekeeping power, the ability to say what is technically possible, reshaping who controls product decisions. Patrick Neeman calls for an agentic standard to stop teams from shipping the same form four different times across different agent surfaces. A separate benchmark from Christoph Hellmuth's team claims to measure AI-readiness across four real design systems, with the evaluation harness published openly.
The sharpest provocation here is buried in the 'Make Me Think' section. A post by Sean Goedecke asks what a human engineer does that GPT-5.6-Sol or Claude Opus 5 would not do in the same role, given that writing code now costs roughly one hundred dollars a month. That question, alongside Anamol Rajbhandari's piece on synthetic identity and Sophia Hee's account of using Claude to build a tool she could not code herself, makes this edition worth reading in full, not skimming.
[READ ORIGINAL →]