On — Nov 14

Google Antigravity rolls out Dynamic Workflows. With /run-workflow, the agent writes a Python script that runs parallel and sequential subagents with structured handoffs, and you review the script before it runs and can save it to run again.

Why it spiked. It replaces prompt-driven agent coordination with a script developers can read and rerun.

• @antigravity Announces Dynamic Workflows and the review-then-run flow for the generated script.
• @_mohansolo Says the feature is heavily used internally and the model can decide to write a workflow without the /run-workflow command.
• @iCleanAI Summarizes the rollout: the agent writes the script, you review it, and you can save it to re-run the same flow.

Anthropic updates Projects in Claude Code: the coordinator that runs tasks across parallel threads now answers questions in the project chat and reports thread status, and /btw and /fork work inside projects. Claude Code on desktop also gets custom themes built on mods.

Why it spiked. Anthropic engineers posted how they use Projects day to day, and a user's theme idea shipped as a feature.

• @ClaudeDevs Announces the more proactive Projects coordinator and status updates on ongoing threads.
• @ClaudeDevs Explains that /btw opens a side thread and /fork starts a new thread from the current one.
• @ClaudeDevs Announces custom themes for Claude Code on desktop, created by asking Claude or starting from the theme-examples plugin.

OpenAI launches ChatGPT Wallet, which lets a dot agent complete purchases with the user's approval. The agent never sees the real card details, every purchase needs a security check, and it is rolling out to Pro subscribers in the US.

Why it spiked. It is one of the first agent payment products built with the major card networks.

• @ChatGPT Announces ChatGPT Wallet, the per-purchase security check, and the US Pro rollout.
• @gdb Says it is built in partnership with American Express, Mastercard, and Visa.
• @Google Adds Google Wallet as a connected app in Gemini for personal finance questions, posted the same day.

DeepSeek makes the features behind its app available in the DeepSeek API. V4.1-Flash (deepseek-flash) supports web search with citations, search during reasoning, file input, URL reading, and safety checks across Messages, Responses, and Chat Completions, with standalone OCR, ASR, and TTS APIs and an open-source frontend demo.

Why it spiked. Developers can now rebuild the full DeepSeek app on the API, and tool calls run server-side.

• @deepseek_ai Announces the API features and links the open-source frontend demo, which can import DeepSeek conversation history.
• @deepseek_ai Lists the built-in tools and says public-facing apps must enable safety checks.
• @deepseek_ai Notes that the extended features carry tool usage fees on top of token charges.

Deepnote researchers release the Agent Initiative Benchmark, which places 12 frontier models as a proactive agent in the Slack of a synthetic company with six years of records. They report that OpenAI's GPT-6.1 Sol and GPT-6 Astra beat Claude Fable 5.2, with Sol at $0.11 per task against $1.87.

Why it spiked. It tests when an agent should step in without being asked, which most coding benchmarks do not measure.

• @jakubjurovych Introduces the benchmark, the synthetic company setup, and the Sol and Astra results.
• @svpino Describes the setup: agents get records across 11 services and face 50 situations.
• @chetaslua Compares cost from the results: GPT-6 Luna at 6.38/10 for $0.007 a task against Fable 5.1 at 7.50/10 for $3.02.

Anthropic invites every US Pro, Max, and Ultra subscriber who was on the Claude Duet waitlist. The research preview counts toward plan limits and includes limit resets: three for Pro and one for Max and Ultra.

Why it spiked. Many users got access to the background agent at once and started posting what it did for them.

• @ahuja_karan Announces the invites, the limit resets, and plans to open more countries soon.
• @mboverell Explains that Duet runs in the background, so users may use up their limits faster.
• @Lucasdickey Says his use of Duet went up two to three times in the past week after voice improvements.

Users speculate about why OpenAI has not released GPT-6.2. Posts point to Fable 5.2, safety or security fixes, or timing around Anthropic's expected IPO; none of these reasons is confirmed by OpenAI.

Why it spiked. Users expected the model this week after hints from OpenAI staff, and it did not arrive.

• @kimmonismus Lists three possible reasons for the delay: Fable 5.2, security issues, or Anthropic's IPO timing.
• @darcyagi Suggests the model may be too capable in some areas for OpenAI to release yet.
• @teortaxesTex Claims OpenAI is holding back and is not behind Anthropic technically.