OpenTeams

AI can write a pull request in seconds. It still can't answer who's accountable if that code breaks something.

Our engineers took this on directly, working through readability, review standards, and where responsibility sits when AI writes the first draft. The piece keeps the open questions open instead of forcing a tidy answer.

Amelia Thurdekoos captured the full conversation.
Read it here: https://na2.hubs.ly/H06Gt5v0

Running an LLM agent locally can reduce API costs and give teams more control over where data and infrastructure live.

But local control does not automatically make the agent reliable.
Adam Lewis explored this by building a Harvest time-tracking agent with Pydantic AI, then switching it from Claude to a local model. The model understood the request, but still fail on details like dates, project names, and time-entry precision.

Adam shares what he learned about designing agents that remain useful even when the model underneath isn't perfect.

Take a look at the full story: https://na2.hubs.ly/H06w5w40

What I Learned Making a Local LLM Do Real Work Learn what makes a local LLM agent reliable enough for real work: evals, deterministic Python logic, and why a better model beats engineering.

At last month's OSA Community event, Dillon Roach walked through the full local AI stack: model selection, quantization tradeoffs, inference engines, agentic workflows, and where to turn when your local compute hits its limits.
A practical map of what's actually usable today.
Worth a watch whether you're already running models locally or just starting to think about it.
🎬 Watch the full session → https://na2.hubs.ly/H06mCPY0

AI on Your Own Terms: Practical Guide to Local Models w/Dillon Roach As large language models are becoming part of everyday life, more p...

TorchDynamo was built to make PyTorch programs faster, but over time its object model accumulated a lot of ad-hoc special cases to handle things CPython handles through a well-defined slot system.

Guilherme Leobas, Senior Software Engineer at OpenTeams, writes about one part of the effort to fix that: the tp_as_number slot, and how binary operations like | and & are actually dispatched in CPython.

The post walks through the dispatch algorithm, how user-defined types plug into it, and what the new slot-based http://dict.or implementation looks like compared to what Dynamo was doing before.

Read the full post on the OpenTeams engineering blog: https://na2.hubs.ly/H06mKgH0

The tp_as_number Slot and Binary Operation Dispatch in CPython | OpenTeams | AI you own Recently Animesh Jain from Meta published a blogpost about the work we are doing in Dynamo. The gist of it is that we are refactoring Dynamo's object m ...