A developer built a knowledge graph that tells Claude Code exactly which files matter for a given change. Instead of feeding it your entire codebase, it traces the blast radius using Tree-sitter and SQLite. Result: up to 49x fewer tokens on a Next.js monorepo. Token costs are the hidden tax on AI-assisted dev, and this is the fix. https://github.com/tirth8205/code-review-graph

I'm a huge fan of locally run AI for development. I can't imagine how much money I've saved in the last few years developing something against a local model and hammering on it for hours, with no token costs. It's a great way to develop software and save money for you or your organization and the upcoming courses in this group will lean heavily on that. Here is a great tool to see which models you can use on your local machine: https://www.canirun.ai/ Check it out, let me know what you think!

Bainbridge's "Ironies of Automation" maps onto AI-assisted engineering with uncomfortable precision Uwe Friedrichsen's two-part series on Lisanne Bainbridge's 1983 automation research caught attention on HN this week. Bainbridge highlighted a key irony: as automation grows, the remaining human tasks become more complex. Yet, people invest less time in these tasks and, as a result, lose skills when they’re needed. With AI coding agents, relying on Claude Code to write and modify code means you understand your codebase less. Anthropic researcher Margaret-Anne Storey recently described this as "cognitive debt." It refers to the gap between what teams' AI-assisted code achieves and what they truly understand. Steve Yegge mentioned he limits his coding with AI to four hours daily due to cognitive fatigue from managing AI output instead of creating it. (Published December 2025; resurfaced on HN this week.) Why this matters: This is not just a theory; it’s a looming risk. Teams rapidly developing AI-assisted systems are accumulating cognitive debt. This could cause unexpected outages, security issues, and difficulties diagnosing failures. The key takeaway for engineers: understanding AI-generated code is essential, not just optional cleanup. Asking the LLM for clear explanations alongside technical output can help reduce these risks. Click here to read more → https://www.ufried.com/blog/ironies_of_ai_2/

This community is returning, but with a twist. We’re moving from general AI use to structured AI engineering career transition. If you're a developer who wants to move into AI roles, you're in the right place. Stay tuned.

Here are the news members from the last couple weeks. Feel free to follow/connect with each other. We've now been around for one month! Thank you all for signing up, this is a part of our "soft launch" where we aren't advertising too much or recruiting yet, just building up content and kicking things off soon. Stay tuned for lots of cool stuff coming down the pipe. Remember since you signed up when you did, you'll never have to pay for membership here if we switch to a paid model. Tell your friends! @Ben Haynes @Philippe Benitez @Mary Killelea @Shay Russell @Dhimiter Mulla @Marcin Osuch @Verena Hladik @Siobhan O'Neill