AI Tooling blogs and newsletters worth following, and what each is good for

These are the blogs and newsletters Briefwire has chosen to cover AI Tooling, each one fetched and checked before it is listed. Company engineering blogs, analyst newsletters and practitioner posts. The receipt is the post it was written in.

Start the 14-day trial See the latest brief

The list

In no ranked order. The note beside each says what it is good for, which is a different thing from how big it is.

Each note describes the publication in its own terms. Briefwire reads what these sources publish, in public, and nothing else.

OpenAI News
Primary source for OpenAI launches, pricing and policy changes.
Google DeepMind
Model releases and research posts that presage Google Cloud AI productisation.
Simon Willison's Weblog
Fast, technically credible read on every model release; useful as a corroborating second source.
Import AI
Jack Clark's policy-and-capability weekly; the standard reference for AI regulation trajectory.
Together AI Blog
Inference-economics posts with concrete cost-per-token numbers analysts can quote.
AWS Machine Learning Blog
Bedrock and SageMaker moves; how the largest cloud is repackaging third-party models.
Replicate Blog
Hosted-inference pricing and model availability; a proxy for open-model commercial demand.
Modal Blog
Serverless GPU economics with published benchmarks and cost breakdowns.
NVIDIA Technical Blog
Hardware and software releases with named customer deployments attached.
Ahead of AI (Raschka)
Technical teardowns of architecture and training method; explains why a release matters.
Don't Worry About the Vase (Zvi)
Exhaustive weekly synthesis of AI news and safety debate; the completeness backstop.
Claude Platform release notes
Anthropic API and model release log; the cleanest primary source for a dated product move.
Cursor Changelog
Weekly AI-coding feature releases; the fastest-moving incumbent's public cadence.
PyTorch Blog
Framework releases that presage hardware and compiler shifts a quarter ahead.
Azure AI Foundry Blog
Microsoft's model-catalog and agent-platform releases; where enterprise AI packaging changes land first.
Epoch AI Gradient Updates
The compute-trends research shop publishing its own data on training runs, hardware and scaling.
One Useful Thing (Ethan Mollick)
Mollick's hands-on capability read each model generation; the reference for what enterprises believe AI can do.
AI as Normal Technology (Narayanan & Kapoor)
The Princeton pair's evidence-first pushback on capability claims; formerly AI Snake Oil.
Hamel Husain
Practitioner essays on evals and fine-tuning that set the vocabulary AI engineering teams use.
Google AI Blog
Official Google AI announcements, separated from the general Google blog firehose.

How the list was made, and what we do with it

Chosen by hand: Every source is here because people who work in AI Tooling say substantive things on it, not because it is large. Enough of them to cover the field, few enough to audit.
Checked before listing: Each feed is fetched and checked when it is added and again whenever the library is re-checked by hand: one whose newest item is more than six months old fails, and one quiet for more than three is flagged but can stay listed. Every source is English-language and public.
Read, so you need not: You do not have to read all of them. Each new post is read, and the claims worth keeping are pulled out, each linked to the post it was written in.
When it is read: Briefwire reads a beat’s sources through the day while a trial or paid account tracks it. A beat nobody tracks may not be read at all, and its page shows the library, not a live feed.
Sent as a brief: The 12 lines that rank highest across the whole AI Tooling beat go out daily or weekly. A claim several independent sources made ranks higher — that counts agreement, not accuracy.

The blogs and newsletters are one part of the beat. Every AI Tooling source, all media — or just the podcasts, YouTube channels or changelogs and release feeds.

Run one of these, or think one is missing? How sources are attributed, and how to reach us.

Blogs and newsletters in other industries

Terms on this page

AI Tooling in 12 lines a brief, each with a receipt.

14 days of Desk, every seat, card on file — cancel before the trial ends and the card is never charged.

Start the 14-day trial See the latest brief