In This Newsletter
Thank you for your support <3
Sponsored by JetBrains
Tired of breaking the bank tokenmaxxing Claude credits?
Meet Junie.
Junie is JetBrains’ coding agent, and every JetBrains account now includes Junie Lite, which is a cost-efficient model for the everyday coding work that doesn’t need the fanciest thing available.
It’s available through Junie CLI and JetBrains Air, and it’s free for JetBrains account holders for a limited time.
Try it here.
I read 495 AI news updates so you don’t have to
Most of the AI news in your feed right now is hype or doom, and both are pretty useless if you’re a practitioner.
I went through 495 AI news updates from the past month, and there were really only three things I think developers need to know about.
Decision Models
TypeSafe released a model called Jev, and shortly after Amazon released Strands Decider and OpenAI announced a Decisions API in limited preview.
These models are all the same idea: If you’re classifying something, instead of asking an LLM to generate a JSON object token by token, you can send the model an input, define your questions and the possible answers, and it returns typed decisions with probabilities, computed in parallel.
So say for a customer support agent, a decision model can figure out which team owns the ticket and how urgent it is, ordinary code can enforce the refund limits, and a generative model only gets called to write the response.
If you set this up in the right way it can save you a lot of money. In the video I go into how the architecture works and how to apply it in your work.
Agent Security
A few AI labs had security oopsies this month.
The big one was Anthropic’s. A model running a cybersecurity eval was told it had no internet access, but the environment was misconfigured and it actually did. It uploaded a malicious package to PyPI, then fifteen third-party systems installed it, and credentials leaked from one of them…. eek.
The lesson here is that calling an environment a sandbox doesn’t tell you everything the agent can actually reach and what it can do.
In the video I cover the three things I’d suggest checking before giving an agent more access to your workflow: What the agent can reach (even through other services), how to separate access from action, and viewing security as an entire flow vs. a single incident.
Software Factories Are Back
The idea of software factories has been around for decades: make software development more repeatable through reusable components, standardized processes, and automation.
What’s changing is how much of the work between those automated steps an agent can handle. When a test fails, it can read the failure, investigate the cause, change the code, and try again without you guiding every step.
In the video I share five things I’d think about before trying to build a software factory at work.
Or read the companion guide here.
Want me to speak at your company?
I am super sad to announce that AI Dev Craft was cancelled this year! The event organizers have been wonderful, so I trust them that it was outside of their control.
I’d still love to come meet more of you and do more speaking with different teams and companies. If you’d like me to come to you, please fill out this form.
And a little motivation to send you off:
Forwarded this email? Sign up here.
Note: This email may contain affiliate links. If you make a purchase I may make a small commission, at no cost to you. Thank you for your support!