CoreWeave announced CoreWeave Forge to connect model deployment, evaluation, and continuous improvement via reinforcement learning rollouts that repeat generate-and-update cycles. It said RL Rollouts helped post-train Nemotron 3.5 Lightning in eight hours using You.com’s web search tools. This shifts post-training toward higher GPU utilization by speeding weight synchronization and enabling cross-region data writes so jobs can feed each other between training rounds.
Offstage is being launched to let Claude Code, Codex, and open-source coding tools run in a separate logged-in macOS account while you keep using your own screen, keyboard, and mouse.
It launches today.
It changes testing and QA workflows by using a second desktop session for simulators, Xcode UI tests, and the app an agent builds, without taking over your primary account.
Hark released Hark Pro, a privacy-focused AI personal assistant that runs a full-screen interface and can take computer actions for users. The wide release is available today, with a free tier plus a subscription tier for heavy users. Users now get an assistant that substitutes for their computer end-to-end and shows web navigation to build trust, while Hark also plans an AI-native device in 2027.
A voice travel concierge was demonstrated for airline apps by wiring spoken requests to backend itinerary, booking, and policy actions using Amazon Bedrock AgentCore plus related AWS services. It streams real-time, bidirectional audio over a WebSocket session. It changes the customer experience by letting travelers tap and talk in the same session to manage travel tasks and optionally escalate to a live agent, rather than navigating screens.
A coding agent from Sakana and Jeff Clune’s lab rewrote its own codebase by adding its own scaffolding and tracking results over about eighty iterations. It improved performance from 20% to 50% on SWE-bench and from 14% to 31% on Polyglot. As a result, the agent gained a persistent history of tried steps, and began validating, generating multiple candidates, and ranking fixes before submitting them.
Ephemeral testing has been proposed as a quality method where an AI agent (or developer) builds temporary software layers on top of existing code, then tests only the resulting layer. The approach throws away the software after testing, using temporary builds as the only test target. It shifts evaluation from the original component to the behavior of the layers produced on top, repeating the process to infer whether the foundation code is reliable.
OpenAI and Ironclad are training and evaluating AI agents on complex contracting workflows to advance computer use for professional work. The article provides no specific dates, benchmarks, or numeric results. As a result, the focus shifts to AI agent performance in contracting-related tasks rather than general computer-use demonstrations.
Hadrian, an Amsterdam-based offensive cybersecurity startup, raised $40M and launched AI-driven penetration testing through its Nova platform with Atlas policing actions. The company says Nova runs more than 100 million attack simulations each day. Hadrian will use most of the $40M for R&D and expand in the United States, aiming to compete in a growing pentesting-as-a-service market.
Serval founder Jake Stauch pivoted from failed neuroscience work into an AI-driven automated service management platform aimed at changing IT from ticket-handling to building tools and protecting data. He received $5,000 to pursue his first startup concept using brain scans to measure advertising efficacy. The story reframes Stauch’s career path as one that leads to applying AI capabilities to fix long-running IT workflow bottlenecks and perceptions.
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
