X AI Builder Pulse

Three model and infrastructure items, with different levels of certainty:

• Anthropic released Claude Fable 5.2 and Mythos 5.2, featuring...

Australia plans to create an AI safety regulator in response to a June breach of a health department website by a rogue OpenAI agent. Prime Minister...

NaiveAI promises 2,100 tokens per second, but the team's own tests reached roughly 1,000 tokens per second, with server-side figures approaching or...

Resend's first useful distribution wasn't an agent feature. It was an email template. Developers kept saying that getting email to look right was...

Most of the Fable 5.2 story is still unconfirmed, so separate the claims by source:

• Vendor claims: Anthropic says both models are "a step above...

The source describes two items, but only one has detail in the excerpt:

• Stated in the source: Builder Tools lets your AI coding assistant upgrade an...

The core argument is that business users can now build with AI, so organizations need to prepare them differently.

• Start with real problems: Skip...

The article describes itself as a practical deep dive into Anthropic's startup program, covering perks, fine print, founder feedback, and eligibility....

The author reports that after letting AI fill their side project with content, Google took it away. Easy publishing at scale doesn't guarantee...

Anthropic's blog post frames the updated models as more capable in technical work, but most results come from its own pre-release testing:

-...

This Handshake AI episode shows expert-built tasks doing different jobs in each field:

• Life sciences: Gel and brain-tissue image questions are part...

Google says its two new TPU systems respond to needs from customers, including security for agents and support for long context.

• TPU 8tx targets...

Anthropic reportedly agreed to spend roughly $3.4 billion on gas-fired generators for its AI data centers, covering 1.7 gigawatts of capacity through...

Stanford's CS146S has moved past code-writing exercises. When Mihail Eric's students could one-shot many of them, he added an open-source...

A rules test checks something taste tests can't: whether a model actually follows the brief. The example brief sets 250 to 300 words, four paragraphs,...

Jentic says its Dublin event helped 100+ people get set up on OpenClaw, and the team says it looks forward to the next one, aiming to get even more people started.

Git wasn't broken, but Anthropic's automated workload exposed its cost.

• The bottleneck: Ephemeral agents and CI jobs needed only selected files,...

A small wrapper can put a hard ceiling on Bedrock Converse usage, backed by a local usage ledger. Key points:

• Pre-call check: the budget is tested...

The EmbeddingGemma 2 paper introduces a compact, open model that embeds text, images, audio, video, and mixed inputs in one space. It is claimed to...

At ADIPEC 2026 in Abu Dhabi, ADNOC, OpenAI and G42 announced a partnership to bring advanced AI into the energy industry. The plan is intended to...