A frontier model got pulled in 72 hours. Tech & AI Weekly by Eli, June 13 2026

Jun 13, 2026

Welcome to my very first newsletter!

Keeping up with AI is exhausting. The field moves so fast that taking a weekend off feels like a career risk, and I was tired of waking up to forty open tabs I would never actually read. So I did the only reasonable thing a developer does when something annoys her this much: I automated it. Now the news finds me every morning, sorted and ready.

And it worked so well that keeping it to myself started to feel a little selfish. So here we are. Every Friday I’ll send you what actually mattered this week, with my honest take on it. Not a pile of links, the why behind them. Because anyone can collect headlines. Understanding them well enough to form an opinion, and to grow in your career, is the part that actually moves you forward. That is what I want to do here, together with you.

If someone forwarded this to you and you want in, you can subscribe here. I’m really glad you’re here.

AI & GenAI

The US government ordered Anthropic to pull Fable 5 and Mythos 5, three days after launch. THE story of the week. The model is out of circulation now, but the lesson for builders stays: a frontier model can vanish in 72 hours, so think hard before you build your whole product on a single one. Read more
GPT-5.5 and Codex are now generally available on Amazon Bedrock. GPT-5.5, GPT-5.4, and Codex landed on Bedrock at OpenAI’s direct pricing. Combined with the story above, the takeaway writes itself: keep your options open and your model layer swappable. Read more
Model triage is officially a skill now. Same coding task: Claude Fable ran about $9, GPT-5.5 ran $1.50. That’s 6x for the same work. Picking the right model per job went from an afterthought to an actual line on your budget. Read more
WebMCP hit Chrome Origin Trials (Chrome 149). Sites can expose real JS functions and forms to in-browser agents, so agents finally stop playing charades with the DOM. It’s the jump from “agent guesses your UI” to “agent calls your API.” Read more
Azure Container Apps Sandboxes (preview) run untrusted agent code in microVMs. Sub-second start, deny-by-default egress. Not glamorous, but it’s the thing that stops the code your LLM just made up from wrecking your environment. Read more
Stack Overflow built a version for coding agents to ask questions. The site that judged your questions for a decade now has a section for the bots. Poetic, honestly. Read more

Cloud & DevOps

AWS introduced CDK Mixins for composable infrastructure. If you write CDK (and I basically think in CDK at this point), it’s a cleaner way to share patterns across stacks instead of the copy-paste ritual we all pretend we don’t do. Read more
The Terraform MCP Server finally hit 1.0 (GA). If you swore this already existed, you’re right, it’s been in preview for the better part of a year. What’s new is the stable release: agents talk to the Terraform Registry directly, so “write me a module” stops being a hallucination generator. If you live in Terraform, it’s safe to lean on now. Read more
Slack moved 700+ jobs off SSH to a REST architecture. A sharp read if you’re trying to get SSH out of your pipelines at scale. Read more

Data & ML

Kimi K2.7-Code: an open-source coding model with better token efficiency. Worth a look if you’re cost-conscious (see model triage above, your wallet remembers) and want an open option. Read more

From me this week

My first YouTube video of the year is live! “Stop AI Hallucinations With These 5 Techniques” on the Developer channel. Five practical ways to keep your agents from confidently making things up, part of my hallucinations series. Watch it here.
AWS Agent Toolkit: Stop Your Coding Agent Hallucinating APIs. This one’s my favorite of the week, and I’ve actually tested it: it’s good. It’s an MCP server, so it plugs into whatever coding assistant you already use. Instead of letting your agent guess API calls from stale training data, it gives the agent live access to 15,000+ AWS API operations (GA since May 6), so it stops inventing calls that don’t exist. Read more
Detect AI Agent Hallucinations: Zero-Shot Methods. Catch hallucinations without training data. The whole series lives on my dev.to if you want to go deeper. Read more

Coming up

Next week I’m recording a podcast I’m so excited about. I’m keeping the topic under wraps for now, but trust me, it’s a good one. Keep an eye out, it’ll be a nice surprise soon.
Codex Community Meetup, San Francisco (June 16). A hands-on evening on agentic coding workflows, if you’re in the Bay Area and want to talk to people building with coding agents in real life. Details

I hope you enjoy reading this as much as I’ll enjoy putting it together every week. Let’s stay informed together, and learn something new along the way.

See you next Friday,

Eli

Dev.to · LinkedIn · GitHub · Twitter/X · Instagram · YouTube