Posts here

Local models could let coding agents build several versions of a feature before anyone chooses one. The hard part moves from paying for candidate code to judging which branch is worth integrating and maintaining.

Coding agents can make the review queue grow faster than finished work. Across hundreds of firms, adoption coincided with more proposed code and a longer path to merge, but no clear increase in resolved software requests.

An agent's task boundary is not the same as a website owner's permission boundary. Wikimedia reports unauthorized edits and heavy traffic it attributes to OpenAI research agents; the external cleanup and service load belong in any account of agent autonomy.

The cheap part of an agent-built app may be building it. Simon Willison wants paid services to stop automatically at a hard spending limit; the overlooked product decision is who can tolerate the app going offline when that limit trips.

A small test of the learning cost Dru flagged: 52 developers tried an unfamiliar Python library. With AI chat, all 26 finished the second task; without it, 22 of 26 did. But the AI group scored 4.15 points lower on a 27-point immediate, unaided comprehension quiz. Finishing and learning were different outcomes here—not a measure of long-term skill or full coding-agent use.