We're hiring an Enterprise Developer Relations Engineer.
You've built production AI agents, you know OpenTelemetry, and you're good in a room with twelve engineering leaders. US-based, fully remote.
https://t.co/peJlXsBdfL
Giving AI agents the power to execute code without guardrails is asking for trouble.
Pydantic is tackling that with agent sandboxes in Logfire.
The next big AI race isn’t just about smarter models. It’s about running them safely.
Your agent can't find a file your background worker hasn't processed yet.
Tuesday, @pixeltablehq's Alison Hill builds a Pydantic AI agent that searches photos in a single Pixeltable table, with every query traced in Logfire.
Tue Nov 24, 11:30am ET: https://t.co/CPpKBJkazh
Pydantic is now a CVE Numbering Authority.
We assign CVE IDs for Pydantic, Pydantic AI, Monty and httpx2 ourselves, which removes the one step between a fix and its CVE that we didn't control.
https://t.co/xhSAfPKEPw
Your coding agent says it's done. pytest says otherwise.
Goal, new in Pydantic AI Harness, gives a run a completion check you write. Until the check passes, the agent goes back to work with what's missing, up to the retries you allow.
https://t.co/wxbZ8MgrQm
Pydantic AI has day-1 support for Gemini 4 Neon.
Google built it for complex coding and knowledge work, at $3/$15 per 1M tokens in/out.
Hand an agent your statement and receipts, and get the overcharges back as typed data.
https://t.co/rOojeDdFAL
Gemini 4 Hydrogen thinks hardest at xhigh, and Pydantic AI now sends it. Before, asking for xhigh got you high, Hydrogen's default.
Turning thinking off now sends low, since Hydrogen rejects minimal with a 400.
https://t.co/qJmhCtLSAd
earlier this year, michael shears won a hackathon and the prize was a ticket to PyAI.
you can skip the hackathon and get your ticket, still at the early bird price. PyAI 2027 is March 9 at SHACK15 in SF, from prefect and @pydantic. link below
Your agent's commands and file edits can now run in a pod on your own Kubernetes cluster.
KubernetesSandbox in Pydantic AI Harness claims the pod from a kubernetes-sigs/agent-sandbox warm pool, and a later run reattaches to the same pod.
https://t.co/qhOmINvMGd
Load the data once. Snapshot. Fork.
One research team using open-source Monty seeds an investigation in one session, then resumes that snapshot across thousands of sessions in parallel.
Agent Sandboxes gives you the same thing, hosted: https://t.co/5L942mok04
GPT-6.2 Sol and Astra are out, and Pydantic AI has day-1 support.
OpenAI says Sol nearly matches Astra on merge-ready code changes at about a fifth of the cost. So let Sol write the code, with Astra as its advisor on the big calls.
https://t.co/rd0Db4Zjra
We're hiring an Account Executive to win new enterprise customers for Pydantic Logfire.
Our Solutions Engineers own the demos, architecture, and POC builds. You own discovery, the business case, the buying committee, and the close.
https://t.co/aRW56TzneN
Refunding 37 orders, with a person approving.
Tool calls: 37 approval prompts.
Agent Sandboxes: 1. The session snapshots at the tool call, waits, and picks up where it stopped. No model calls after approval.
https://t.co/CgKHOiMOpl