Taste in the Loop

Adam Bender gave a great talk at Google I/O this year about the future of software engineering, “software ecology”, and all the things that will break due to AI agents “10x-ing” the throughput of our ability to write code. Definitely worth watching.

MiniMax released M3, their newest LLM, yesterday (June 1st). MiniMax claims M3 benchmarks competitively with Opus 4.7 and GPT 5.5, will be open-weight soon, and comes at a fraction of the price of those other models — only $0.60 per million input tokens and $2.40 per million output tokens (when context is under 512k tokens). Historically, MiniMax’s M-series models have been really good (for their size), but don’t perform as well in practice as the benchmarks suggest (which is common for Chinese models). This one is worth trying out for your own use cases.

A great post by Armin on the challenges of dealing with AI slop while using Pi (Earendil’s coding agent) to build Pi. This post also covers the challenges associated with running a large open source project at a time where a lot of the people submitting issues are submitting bad issues full of slop. This is important because wielding coding agents effectively requires writing good issues, which we’ve covered in the past.

Jamie has a very practical post on how to move faster. It’s an older post (from 2021), so it focuses on the things you can do as a human to move faster, rather than what you can do with AI coding agents. His discussion on the limits of human attention and the cost of multitasking is especially relevant today given the push for people to frantically alt-tab across multiple coding agent sessions.

Consequently, this is one of the reasons why background coding agents are so valuable; they take you out of the loop and allow you to focus your attention on the most important task at hand, rather than trying to spread your attention across N tasks concurrently.

LLMs are extremely powerful (and useful) tools, but they shouldn’t be used in ways that remove our own voice or deprive us of human social connection. Shawn writes a beautiful and heartfelt essay on the pitfalls of doing that.