Company@X

Signal of the Day

OpenAI has unveiled the first performance results for Jalapeño, its custom AI inference chip, claiming a major power-efficiency leap alongside higher throughput and lower latency. The company plans to begin deploying the silicon in its production compute infrastructure by the end of 2026, marking a critical step in a multi-generational custom hardware roadmap to scale its services.

Key Announcements

Skild AI · Source The robotics startup has launched S1, a new foundation model capable of executing long-horizon tasks (spanning over 10 minutes) from a single video prompt without any task-specific fine-tuning. S1 demonstrates general-purpose robot intelligence by executing complex physical manipulations far outside its training distribution, such as preparing pancakes after watching a single demonstration. This breakthrough represents a major milestone for in-context learning directly on physical robotic hardware.

Adaptyv Bio · Source Adaptyv Bio has raised a $40 million Series A round to construct high-throughput automated wet lab infrastructure—a “biological gigafactory”—for agentic biology. In a joint effort with Anthropic, the company verified that Claude can autonomously design functional, novel protein binders from scratch, which were then physically synthesized and tested in Adaptyv’s automated labs. This partnership establishes a closed-loop engineering system where AI agents design, execute, and iterate on biological experiments.

Waymo · Source Waymo is expanding its fully autonomous ride-hailing services internationally, announcing plans to bring the Waymo Driver to Munich, Germany. Mapping and testing vehicles will arrive in Munich over the coming weeks to begin laying the groundwork. The deployment targets a commercial, fully autonomous ride-hailing service launch in Munich in late 2027.

AWS · Source Amazon Web Services has launched AWS Agent Registry under Amazon Bedrock AgentCore to manage and govern enterprise AI agent deployments. The registry allows organizations to discover, register, share, and reuse AI agents, tools, and agent skills across internal teams. This product directly addresses the operational challenge of “agent sprawl” and technical debt as enterprise autonomous workflows scale.

OpenAI · Source OpenAI has upgraded ChatGPT Work with the capability to autonomously log in and interact with third-party web and mobile sites. The agent operates securely within a browser environment, shielding user passwords from the model while carrying out end-to-end tasks like booking appointments or preparing utility registrations. This capability transitions ChatGPT from a conversational assistant to an active browser automation agent.

Apodex AI · Source Apodex AI has launched its Apodex 1.1 model family, introducing frontier-level intelligence designed to coordinate asynchronous agent teams. Alongside the models, the startup open-sourced FrontierAgent, a locally deployable research workbench that lets users break down complex tasks and manage multiple agents in parallel. The release includes a high-capacity frontier model, a local open-weight version called “Apodex 1.1 mini,” and API access.

Floot · Source The development platform launched an integration enabling users to build, debug, and publish live applications directly inside Claude and ChatGPT on web, desktop, and mobile. Floot automates the underlying infrastructure, bypassing Git, Vercel, and database configurations to publish directly to a custom domain. The service is free to start and does not charge separate AI credits, instead running entirely on the user’s existing Claude or ChatGPT subscription.

Also Noted

• Eight Sleep (Source): The consumer sleep-tech company is seeking FDA clearance to turn its smart mattress data into a covered medical service for sleep apnea.
• Cursor (Source): The developer platform has permanently increased included Grok usage limits in its editor to support growing demand following xAI’s Grok 4.6 launch.
• Rasyn Lab (Source): The research group open-sourced Synthon 350M, a retrosynthesis model that achieves an industry-leading 66.23% top-1 accuracy in identifying candidate chemical reactants.
• Greptile (Source): The AI code review tool rolled out knowledge bases to maintain running histories of past bugs, boosting bug detection rates by 12.2%.
• Bluejay Labs (Source): The audio data lab open-sourced MIVAS Bench, a voice AI evaluation suite featuring 8,600 simulated conversational tasks to benchmark vocal agents.

📊 I can compile a comparative architectural breakdown of the developer-focused AI toolchains and environments launched today—such as Apodex, Greptile, Floot, and OpenTag—to evaluate their differences and deployment models.