🦸 Community Superhero

Our community superhero this week is Andrea Griffiths, senior developer advocate at GitHub. Andrea has a sharp focus on developer experience, sharing her expertise in everything from workshops to conference talks to stellar content. She also recently started publishing a newsletter she calls Main Branch, which talks about everything she’s doing, what tools she’s using, and cool articles from around the web. Thank you so much for everything you do, Andrea!

💯 Spotlight

I’ve been liking what I’m seeing with the research and experiments lately with Jev. The reception has been outstanding and now that we’re seeing numbers from practitioners roll in, it’s a pretty convincing sell to start adding it to our AI workflows. I really liked how Boris Tane and Vi Tran broke down their usage of it and how it has helped both cut costs and reduce latency in their app. I didn’t realize it had so many different use cases, so it’s super cool to see the evaluation criteria and exactly how to build them yourself.

🔥 My Favorite Content

If our spotlight wasn’t enough to convince you to try out Jev, maybe Jeremy Daly’s post will. Jeremy takes a much more scientific-feeling approach, with deeper explanations and a more holistic view of what decision models do to your software. His thesis in the post is that you should stop asking the reasoning model to decide everything, and he backs it up with numbers and thorough proof. Reading this gave me a much better understanding of why I’d want to implement Jev outside of faster execution and lower operational cost.

Speaking of making systems work better, I found a practical one from Elena van Engelen-Maslova about how she set up Kiro to work optimally for her workflow. She goes through quite a few components of it I was only particularly familiar with, like the importance of .kiroignore, but she also talks about the effectiveness of many of the core features of Kiro. I really like this post because she keeps circling back to something engineers always have a hard time with: keep it simple.

But what about making your agents work better? Better is a subjective term, but what if there was a way you could make it objective? Well, you can’t. But I tried 😅. I published a blog post last week claiming agent memory needs more than vector search, which I stand by. The article goes through a benchmark I did that measures relevance of retrieved memories in a RAG pipeline under different circumstances. My goal was to see if rerankers were worth the added latency if they increased relevance substantially. It’s a pretty good read, and I did not reach the conclusion I thought I would.

Putting everything we’ve talked about together culminates in a story pretty close to what Kaylyn van Norstrand published last week. She wrote about how she cut agentic AI costs by 44-85% by swapping to a harness from Vercel. Don’t worry, Kaylyn explains what a harness is, but also talks about how the one she opted for handles compaction, memory, and automatic model switching. I think coupling this method with the decision making capabilities of a model like Jev will make things faster, cheaper, and substantially better than anything we’ve seen yet in 2026.

💡 Tip of the Week

If you needed more of a reason to try Jev this week, apparently Cloudflare saw the appeal and just released their own open-sourced version, Clef, last week that runs even faster 🤯 It’s a little bit more expensive, but if latency is your thing, maybe Clef is worth a shot instead?

Last Words

I’ve been really enjoying watching our industry start converging on strategies and methodologies for coding with agents and building the agents themselves. It’s great to see the focus shift from “let’s just make it work” to “let’s make it work properly,” which is exactly what’s happening with Jev and other decision models. I am excited to see what the rest of the year brings, and looking forward to an even crazier 2027.

That’s my take on the week, but what’s yours?

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