What the people actually building AI said today. One page — a 2-min read.
Google announced another Gemini model just two weeks after Argon, while Vercel pushed to give coding agents more control over CI. Elsewhere, the practical uses were specific: legal workflows, state agency development, and creative tools that keep people in the process.
Gemini 4 Carbon rolls out to cyber defenders first
Josh Woodward says Gemini 4 Carbon is rolling out to cyber defenders before a wider release, two weeks after Gemini 4 Argon. Gemini also added a Strava connection that turns workout logs into a personalized coaching plan.
Vercel gives coding agents control over CI runs
Vercel CI now lets agents inspect, abort, trigger, and replay runs from the CLI. Guillermo Rauch says Turborepo can skip and cache work across a project's dependency graph, and describes model aliases that steer traffic without tying teams to a fixed version.
Claude for Legal adds ready-made workflow artifacts
Claude for Legal now has a gallery of artifacts for contract review, regulatory horizon scans, and citation checks. Teams can start with an example, add their own documents and connectors, then share the result across the business.
Fred Graver's case for AI as a writer's room
Fred Graver says creators should first write down what they know and want to make, then use AI to test ideas, production choices, and ways to reach an audience. “You have to start with uniquely you, what you know.” He sees distribution as the hardest remaining problem for independent storytellers.
OpenClaw's web UI now works offline
Peter Steinberger says OpenClaw's web UI works offline and uses WebSocket compression and reconnection logic. He says the changes let him keep working through spotty airplane Wi-Fi.
Aaron Levie expects agents to drive far more model usage
Aaron Levie predicts agents will consume 1,000 times more tokens as they enter workflows, process data, build software, and secure enterprises. He argues that demand will require stronger closed and open-weight models, along with better security and efficiency for large workloads.
GPT 6.1 Sol leads the new ViBench prompt-to-app benchmark
Amjad Masad says GPT 6.1 Sol leads ViBench 2.0, a benchmark for turning prompts into apps. He shared the result without scores or a breakdown of the other models.
Source data comes from the open-source project follow-builders by zarazhangrui, released under the MIT license. Summaries are generated by an LLM from that project's public feeds, and the summarization prompts are adapted from it. Every item above links to its original source.
Summaries generated automatically. Read the original before relying on any claim.
