Cursor releases the Agent SDK, generally available today, letting teams define agents in code and deploy them to Cursor infrastructure with one command, agent-sdk deploy. Cursor says a team built a knowledge-base reviewer in about a day, and PGA Tour engineer Michael Gyarmathy says the tour uses it to build its software factory.
Hayden Smith released Floorplan 0.6.3 on Nov. 2, 2026, a plugin that has Claude Code or Codex draw an interactive, plain-English map of a codebase. The project says the release finds more endpoints and links them correctly. It runs on the user's existing Claude or ChatGPT plan, with no API keys.
Flamerix AI releases v4 of no-agent-summary.js, a Bun --preload hook that blocks Claude Code's agent_summary timer, which re-sends background agent transcripts to fill status lines. Version 4 fixes Windows support: the earlier gate never matched Windows entries or backslash paths. Flamerix says the query consumes 15 to 20 percent of tokens.
OpenAI tells users it will merge ChatGPT's Chat and Work tabs, starting with select users before a wider rollout. Both ChatGPT and Codex will draw tokens from one shared pool, OpenAI says, which could mean more frequent waits for resets for some heavy users. Anthropic combined Claude's chat and Cowork functions in September.
Anthropic reports about $17 billion in third-quarter revenue and $895.4 billion in aggregate purchase and lease commitments, mostly for compute and infrastructure, in an amended Form S-1 filed with the SEC. The filing updates its IPO prospectus with third-quarter results. Anthropic posts a $23.29 billion quarterly net loss.
Anthropic tells investors in a revised S-1/A filing that it may imminently meet its Responsible Scaling Policy definition of dramatically accelerated progress attributable to AI, the threshold it calls recursive self-improvement. The company says no current model has crossed that line and pledges to keep humans in frontier model development. The filing is dated November 2.
NVIDIA released Cosmos 3 Nano recipes and checkpoints that generate synchronized camera video and LiDAR for autonomous vehicle, robot, and surgical scenarios. The company says Cosmos NIM microservices deliver roughly 2 to 3 times faster inference on H200 GPUs through quantization, speculative decoding, and KV-cache-aware routing. Cosmos3-Nano generation needs 58 GiB in BF16 and 44 GiB in FP8 with offloading.
Framer rebuilt its Agent harness to execute JavaScript while the model still streams it, compiling complete statements into cells that run in a Web Worker. Framer says the change cuts tool steps by 43% and cached input tokens per message by 39%, with runs up to 33% cheaper and 20% faster. Cancelling generation on a failed assumption avoids output-token charges.
GitHub Next researchers propose enclaves, a single-use sandbox that limits what an agent reports about a private repository. GitHub Agentic Workflows, generally available as of last week, uses a trusted broker to check responses against a schema and a per-run information budget: a confidential repo allows 8 bits. The team says this blocks the GitLost private-to-public leak.
Anthropic releases message threads in public beta on the Claude Platform, letting agents send only new messages per turn instead of resending full history. Anthropic says requests of 500k to 1M tokens cut median time to first token by about 35%. Threads expire five minutes after the last request, and token usage is unchanged.