"What I want most: software that helps me express, formalise, and include my opinions rather than enforcing its opinions on me."
In the 3rd part of the Box Model series, @summerscope.bsky.social talks about the design process, intuition, and cargo cult prompting.
pydantic.dev/articles/lie...
Link to the talk: www.youtube.com/watch?v=tnFD...
A model discovered a contradiction between the Bible and biology textbooks. It had access to a lot of compute so it consumed an entire data center's worth of it, debating the Immaculate Conception with itself.
This talk is 15 mins of real production AI failures and what it took to debug them.
@summerscope.bsky.social 👆
"The best navigation is the kind nobody notices."
On Part 2 of Lies, Damn Lies, and the Box Model, @summerscope talks about the design process behind Logfire's sidebar.
The trade-offs, and what "directionally correct" means in practice.
Part 3 (and final) out tomorrow.
AI observability pricing is a lot.
We modeled three scenarios across Logfire, Arize AX, Langfuse, and LangSmith. Same assumptions, real numbers.
At 5 users and 50M spans/mo, Logfire is between 8 and 40 times cheaper than competitors.
pydantic.dev/articles/ai-observability-pricing-comparison
It's 2026 and AI can write code from a napkin sketch.
Still, our lead design engineer spent an hour debugging a scrollbar with help from Claude Code.
A story about CSS, containing blocks, and the things AI can't really see - part 1.
pydantic.dev/articles/lie...
We have been in the telemetry industry for a while now with Logfire. pydantic.dev/logfire
Proud to be the platform that unifies LLM & full-stack traces. That you can query like a database.
I'm looking forward to returning to streaming at 9:30 AM UTC tomorrow 27th March.
Please join myself and Petr Victorin as we begin a project to work on CPython's `str` implementation - targeting optimizations when exchanging data with Rust UTF-8 strings.
www.youtube.com/live/fXeqeFX...
In a nutshell. that's what he says. Plus "nothing beats SQL for querying telemetry data".
"How do agentic coding tools work? I have no idea. And you probably also don't."
That was @mitsuhiko.at opening his talk at PyAI.
What follows is one of the most useful 20 minutes on working with agents we've ever seen.
youtu.be/8RHYyRUxVrA
"How do agentic coding tools work? I have no idea. And you probably also don't."
That was @mitsuhiko.at opening his talk at PyAI.
What follows is one of the most useful 20 minutes on working with agents we've ever seen.
youtu.be/8RHYyRUxVrA
Pydantic models were using more memory than expected. We dug in with Memray, found the bottleneck, and replaced it with a Rust bitset.
55% less memory. Faster validation.
Full deep dive on the new blogpost pydantic.dev/articles/pyd...
What a night! Thanks @surrealdb.com for hosting us! We had a blast.
Last week we hosted our first conference, PyAI. The community showed up, we had an incredible event.
Here are some of the highlights of the day.
youtu.be/Sbut6lWudzY?...
Curious how to keep an agent’s output correct as data changes, memory grows, and systems scale? Join Tobie Morgan Hitchcock and guest speakers @sa.muelcolvin.com (@pydantic.dev) and @badmonster0.bsky.social (@cocoindex.bsky.social) at our San Francisco meetup.
Register now. 👉 sdb.li/4sryXOz
A few weeks ago our founder announced Monty, a minimal, secure Python interpreter written in Rust, for running code written by AI agents. People got excited. @sa.muelcolvin.com just wrote up the full story on why he's excited about it and you should be too.
pydantic.dev/articles/pyd...
We asked Adrian and Peter from @pydantic.dev what would block the Rust-accelerated Python trend from growing even more... Their answers:
Pydantic AI Gateway is now in open beta 🚀
One API key for OpenAI, Anthropic, Google, Groq, and Bedrock. BYOK and Built-in providers. Real cost limits. Built-in observability. Zero translation delays. Free during beta.
pydantic.dev/ai-gateway
Some personal (very belated*) news, I've moved to the bay area!
Let me know if you're in SF or Palo Alto and want to get coffee.
Also, last reminder that we have the inaugural Py AI meetup with @pydantic.dev in SF on Tuesday. Register now: luma.com/gjtfecc7
* I actually got here in the summer.
Workflow graph for the deep research agent, showing massive parallel processing (fan-out).
I added two new demos showing durable agent patterns (parallel vs sequential) with @pydantic.dev AI + @dbos.dev:
- Deep Research: large fan-out parallel workflows
- Twenty Questions: long sequential subagent chaining
Both are long-running, so durable execution is necessary to survive crashes.
SentinelOne just announced their Purple AI MCP Server at #OneCon. Now, you can access SentinelOne’s security platform with your Pydantic AI MCP agent!
Check out their GitHub repo: github.com/Sentinel-one...
A screenshot of using Logfire + DBOS + Pydantic AI agent.
I wrote a tutorial on integrating DBOS with Logfire. It's super easy to set up (kudos to the @pydantic.dev team!) -> just a few lines to configure the exporter and your Logfire write token. Once it's running, you'll get a unified view of your app with logs + traces in one place.
I'm teaming up with @sa.muelcolvin.com to launch a new meetup series, Py AI.
Our first one is November 11 in San Francisco with talks from @pydantic.dev, @modal-labs.bsky.social, ReductoAI and @fastmcp.bsky.social.
Hosted by our friends at pebblebed.com
RSVP here: luma.com/gjtfecc7
Just putting this here. Not much more to say.
@pydantic.dev
New social icon for @opensourcepledge.com spotted on
@pydantic.dev's website. Lookin' sharp! 👀
