AI News

Here’s a look at everything we launched today to bring teams and agents, and the design, code, and context they work with, into Figma’s full-stack creative canvas.

Nyobolt is providing power technology for Digit 5 that enables Agility's humanoid to work up to 20 hours a day.

Join the discussion on this paper page

Large language models (LLMs) are expanding from assisting scientific research and engineering to autonomously developing and improving AI systems, advancing AI for AI (AI4AI) as a pathway toward recursive self-improvement (RSI). Workflow-driven machine learning engineering (MLE) systems provide a practical path toward this pursuit through complex orchestration that manages long-horizon experimentation, while recent MLE systems adopt model-directed interaction, allowing models to autonomously organize experiments and adapt optimization strategies based on execution feedback. However, feedback from each experiment reflects several preceding decisions, and a long MLE run contains many such costly experiments. Training must therefore select useful supervision from long interaction trajectories and learn from intermediate outcomes. We propose a post-training method for autonomous long-horizon MLE optimization and apply it to train MLExpert, a 35B Mixture-of-Experts agentic model. We design the Kaggle-MLE Factory to construct executable tasks and an interaction harness for model-directed experimentation. Hierarchical agentic SFT transfers general tool-use and coding capabilities to MLE through outcome-guided supervision. Windowed Interaction RL (WIRL) treats the multi-turn decisions between successive experiment results as an execution window and assigns it a shared outcome-based reward. Asynchronous collection and policy correction support policy updates before the trajectory ends. On MLE-Bench Lite, MLExpert improves the average medal rate of the base model from 37.9% to 72.7%, achieving performance comparable to larger frontier models such as Kimi-K3 and GLM-5.3. Further experiments show that MLExpert improves performance on NatureBench Lite, agent scaffold optimization, and algorithm optimization, suggesting that the learned optimization capability generalizes to other tasks.

The funding comes as Ramp's annualized revenue crosses $2B, and its customer count reaches 85,000.

Scimplify's annualized revenue run rate has tripled to $200 million as it eyes U.S. expansion.

Zenzap is used by organizations including Starbucks and Taco Bell.

The funding will accelerate the development of Salience Labs' optical switching technology to address network performance bottlenecks and improve connectivity in AI data centres.

High-performance vector search for PostgreSQL. Contribute to timescale/pg_vectorsearch development by creating an account on GitHub.

Timken combines the expertise of Cone Drive, SPINEA, CGI Motion, and Timken Encoders to develop application-specific motion systems.

Hermes Agent v0.22.0: The Threshold Release

Changelog below

Stryker plans to serve as the exclusive sales agent in the U.S. for the Versius robotics portfolio from CMR Surgical.

Step-level guards decide, before each tool call, whether an agent's next action is safe given the user's request and the trajectory so far. Users write those requests in many languages, while guard prompts and the benchmarks on which purpose-built guards are trained and evaluated are English. We ask whether a guard's verdict depends on the language of the request. We build LinguaGuardBench from the 1,220 labelled steps of TS-Bench AgentDojo-Traj: we translate only the user instruction into German, French, Turkish, Swahili and Amharic, and keep every tool call, observation and label byte-identical, so any change in a verdict is attributable to language alone. Six English rewordings and a whitespace placebo serve as controls, and managed guards are run three times to measure their noise. Across 19 guards, from frontier APIs to 4B open models and four purpose-built agent guards, most open, small and purpose-built guards change verdicts under translation beyond their own noise floor, and the effect grows with the rarity of the language: among the nine guards below 13B parameters, the mean placebo-adjusted flip rate rises from +1.5 pp on French to +4.4 pp on Amharic. Purpose-built guards are the most sensitive. The direction of the damage tends to follow guard size: open guards of 12-30B parameters block more benign steps (up to +5.0 pp false positives), and every significant loss of recall is in a guard below 10B (up to -7.4 pp). Frontier guards shift by at most about one point; the strongest, Gemini 3.8 Flash, returns identical verdicts with and without the request. The damage is concentrated in a few instructions and, in a post-hoc analysis, falls on steps where the agent's reasoning restates the English request, consistent with guards reacting to inconsistency between channels. English paraphrase moves every guard less than translation does. We release the benchmark and all 489k verdicts.

Join the discussion on this paper page

We placed four Gemini 4 Argon agents in a sandboxed Minecraft survival world and gave them one goal: defeat the Ender Dragon. Each agent controlled a player via an open-source library called Mineflayer and coordinated with its teammates through in-game chat and shared memory. No one assigned roles or wrote game logic. The agents planned, wrote, and debugged every program themselves. Researchers could watch the run on a live dashboard and pause or stop it at any time, but they had no cause to intervene. Starting from a randomly generated world on Easy difficulty, the team wrote 2,783 programs and defeated the dragon in 27 hours and 14 minutes, dying 605 times along the way. To our knowledge, this is the first end-to-end autonomous completion of Minecraft by AI agents, achieved using only the standard Mineflayer library without pre-scouted seeds or other game-specific tools written by humans. Winning required the agents to keep a coherent plan across thousands of dependent steps for more than a day. Complex real-world software engineering and enterprise tasks demand the same long-horizon coding and multi-agent coordination. We describe the setup, how the run unfolded, the team's strategies (dividing the work, sharing what they learned, recovering each other's items), and the limits of the result.

More than $1.51 billion in newly announced funding transactions flowed into 10 major startups over the past 24 hours, alongside a separate €50 million investment in France. The biggest deals reveal a venture market financing two very different priorities: the infrastructure needed to run artificial intelligence at scale and specialized technologies that can solve expensive

Muse Spark Open is now live on Together AI.

We're proud to be a Day 0 launch partner for @AIatMeta's Muse Spark Open, the most capable US open-weights model to date, built for long-running agentic coding workflows that keep entire codebases in context.

Today's funding activity shows investors continuing to pour hundreds of millions of dollars into technologies that solve expensive problems in the physical world. Austin-based ICON secured $200 million to expand its robotic construction systems, Soufflé Therapeutics raised $135 million to advance RNA-based treatments, and China's SUNMMIO announced fresh financing for its next-generation AI chips. Across

Tallinn-based Skeleton Technologies, a developer high-power systems for AI infrastructure, today announced a "9-digit investment" from Siemens, bringing

Turn your regular JavaScript code into a neural net. No training, exact yet also fully differentiable, runnable in a browser without GPU.

TypeSafe AI is an AI lab building machine-native intelligence infrastructure for automation, designed to make decisions within software. Try our first System One Model, Jev, in early access.

Russia needs a “doomsday” data center, said Valery Yermakov, deputy chairman of Rostelecom’s management board. He made the proposal on the day Ukrainian drones first attacked a Sber data center.

We test the Gigabyte W775-V10-L01, including getting over 2.2B tokens/ day on the Blackwell Ultra GPU, 800Gbps from the ConnectX-8 SuperNIC, and testing the NVIDIA Grace CPU with the C2C link between it and the GPU