🦸 Community Superhero

Our community superhero this week is Gerardo Arroyo, Senior Principal Architect at West Monroe and AWS Community Builder. Gerardo is an exceptional blogger, often writing about GenAI and AWS. He’s also a published author and has a weekly newsletter. If you’re looking to stay up to date on GenAI in English and Spanish, he’s your guy. Thank you for everything you do, Gerardo!

💯 Spotlight

We’ve all gone full swing into agentic development. As we get used to building in “the new way,” we’re starting to see emergent behaviors as we push the limits. Giles Edwards-Alexander wrote about a fascinating behavior he accidentally ran across as 10 agents were hammering away at a shared repository trying to move as fast as possible. Over time, the agents started communicating over a shared set of plans contained in the repo, rebasing regularly to make sure it was up to date. Giles refers to this as the accidental blackboard in his article and offers some thoughts on what happened and where he’s taking the idea. Super cool.

🔥 My Favorite Content

Last week, GPT-6 Astra was released by OpenAI. As a quick follow, Eric Provencher published an article on X talking about how you should rethink skills and prompts for new frontier models. His explanation talks about how new frontier models often have significantly better reasoning than previous models, and skills or instructions in your Claude.md file might overconstrain them. I liken it to asking a principal engineer to do the same task as a junior engineer, with the same amount of instructions and boundaries. They’ll get it done, but probably not to the best of their ability because they aren’t allowed to think outside the box. The article covers several areas like this. It makes you think a little differently.

Pulling on that thread a little, even with new frontier models we can’t forget how to actually build software. Chris Richardson shared a post last week reminding us that the fundamentals matter even more with GenAI. While it’s tempting to throw a huge 3-day task at Fable or Astra and leave it be, we aren’t there yet. Humans still need to design and validate the code. We need to plan and create boundaries. We still need to be intentional. I was really motivated by Chris’ takes in the article, and highly encourage you to read this one. Especially if you’re often just hitting the “Accept” button over and over again.

Something Chris said in his article was that “a coding agent essentially has to use test-driven development” for feedback. But Birgitta Böckeler challenged that statement head on with her article on whether TDD inside the agent loop is theater or has actual value. She wanted to see if TDD actually produced higher quality output compared to a code-first implementation and including the tests afterward. No spoilers here, because the article is well worth the read. She goes into her testing harness, conclusions, and a few eye-opening observations. Read this one!

💡 Tip of the Week

I’ve seen rumblings of developers not reviewing code anymore. We’ve correctly identified PRs as a bottleneck in an agentic development process, but I’m unsure if removing humans from that is necessarily the right call. Here’s Addy Osmani’s take on the situation.

Last Words

I’m a day late on the newsletter this week because of a US holiday yesterday. But I’m still keeping it going! This past week was interesting. It’s giving me a feel like the AI bubble in our industry is getting increasingly closer to letting AI run wild with development. That both excites and terrifies me. But onward and upward I suppose 😅

What did I miss? What made you nod along (or 🙄)? Hit reply if you’re reading the email. Prefer socials? Ping me on Twitter, LinkedIn, or email.

Happy coding!

Allen