Welcome to this week's Top 7, where the DEV editorial team handpicks their favorite posts from the previous week (Saturday-Friday).
Congrats to all the authors that made it onto the list 👏
@kenwalger built a wine cellar app with an AI agent writing most of the code, then discovered that many of its passing tests were proving nothing at all. Borrowing the idea of positive and negative controls from lab science, they break down four distinct ways a test can lie and the habits that finally made verification hard to fake.
@dannwaneri responds to a viral claim that no human will ever out-code an LLM by asking a simple question: what exactly are we measuring? He walks through a hierarchy of what "better" can mean, from lines per minute to correctness to maintainability, and makes the case that the wisest code is sometimes the code you decide not to write.
@gramli hid a prompt injection inside an ordinary-looking application log and handed it to several AI models to see which ones would abandon their task. The results of this Halloween experiment are a fun and slightly spooky example of how untrusted data can be mistaken for instructions.
@sarvar_04 rebuilt a real AWS cost-triage MCP server from fifteen thin tools down to four outcome-level ones, cutting the calls needed to answer the same question from seven to two. He shares the measured results, the design fixes behind them, and an honest look at where the context savings actually matter.
@cseeman audited ten weeks of her non-coding AI chats, from trip planning to a mysterious cat food smell, to see whether any of it actually saved time. Her takeaway is that the most valuable conversations went past the question she typed and answered the one she should have asked.
@maame-codes was tasked with migrating a server as cheaply as possible during her internship and thought AWS Spot instances were the obvious win, until a single question at standup changed everything. She explains the pets vs cattle distinction and the one question she now asks before choosing Spot for anything.
@esratech built a full agent stack entirely in Dart, with an approval gate that stops the AI from spending money without a person's OK. Along the way they uncovered a framework bug in how approved MCP tool calls resume, and they share the workaround along with their first open source fix to Genkit Dart.
And that's a wrap for this week's Top 7 roundup! 🎬 We hope you enjoyed this eclectic mix of insights, stories, and tips from our talented authors. Keep coding, keep learning, and stay tuned to DEV for more captivating content and make sure you’re opted in to our Weekly Newsletter 📩 for all the best articles, discussions, and updates.
Top comments (8)
Congrats @kenwalger, @dannwaneri, @gramli, @sarvar_04, @cseeman, @maame-codes, @esratech!!! 👏👏👏
How did I not see this?
Well, congrats ya'll!
Thank You Jem ❤️
Thank you, @heyitsjem! Congratulations to everyone on the list. The link shows a 404 page for me too. I also could not find the post on your profile. Can the DEV team check if it is published??
Thanks, @heyitsjem, what an honor.
Congrats to @dannwaneri, @gramli, @sarvar_04, @cseeman, @maame-codes, @esratech as well for great work and continued community contributions.
Congrats everyone! 👏
Looks like my Halloween ghost made it into the Top 7! 👻🤣
Thanks so much for featuring my article! ❤️ And congratulations to all the other authors! 👏
Thank You Jem 🙌🏻