How Merge Mommy scores pull-request risk

#1 ▶️

Build an AI code review agent with Vercel Eve (full tutorial)

How I AI Podcast

Also covered by: @claire vo 🖤, @Claire Vo

#2 𝕏

AI at Meta says Meta released Muse Spark 1.2 in Muse Code and the Meta Model API, with expanded global access. More features are planned, though details and release dates were not provided.

Also covered by: @AI at Meta, @Alexandr Wang, @Alexandr Wang

#3 𝕏

Vercel released the new v0 API, which can be used to ship an app builder, let agents build and deploy apps, and generate apps from a script or CI job.

#4 𝕏

Cognition shared that Devin Outposts can run on Vercel Sandbox to build and test apps in an isolated microVM, with support for Docker, private-network access via VPN, and filesystem snapshots that preserve repositories, dependencies, and build state.

#5 in

Guillermo Rauch shared that one line of code in AI SDK saves 90% or more in DeepSeek v4 Flash AI Gateway tokens.

#6 𝕏

Guillermo Rauch announced Next.js 16.3, describing it as radically more efficient to serve, faster to build, and capable of massive compute and data-transfer cost reductions at scale, with an easy upgrade path.

#7 in

Colin Matthews recapped OpenAI’s recent work involving 10 previously unsolved math problems to highlight verifiability: checking LLM outputs against external truth, such as Lean-verified proofs, so models can iterate without a human in the loop. He suggests building LLMs as search agents that test candidate solutions against verifiers, though creating such systems beyond math and computer science remains challenging.

#8 𝕏

Peter Yang announced /human-review, a free, open-source AI skill under the GitHub account petergyang that opens HTML and Markdown files in a visual editor, with the review loop running locally. Users can directly edit text, resize images, leave comments for an AI agent, and send their edits and feedback to the agent to apply.

Also covered by: @Peter Yang

#9 𝕏

Madhu Guru shared a model-selection playbook: prototype with the best frontier model regardless of cost, validate the user experience, then move production workloads to open-weight and smaller models where possible as they catch up 6–8 weeks later. He advises against starting with the cheapest model.

#10 𝕏

Qwen says Qwen-Image-3.0-Pro is now live on Qwen Cloud.

#11 𝕏

Jeff Dean announced Discovery Loop, a Public Benefit Corporation whose stated mission is to automate machine learning, science, and engineering.

Also covered by: @Sundar Pichai, @Jeff Dean

#12 𝕏

Sundar Pichai announced leadership changes at Google DeepMind: Demis Hassabis will become Chair and Alphabet’s Chief Scientist while continuing to lead Isomorphic Labs, focusing on AGI and scientific discovery. Koray, a 13-year Google DeepMind veteran, will become SVP overseeing model development, research, and the Gemini app and developer teams.

Also covered by: @Demis Hassabis

#13 𝕏

LlamaIndex 🦙 recapped that across three GPT generations, parsing accuracy gained ~24 points while cost per page 4x'd, and the newest frontier models still trail specialized parsers.

#14 𝕏

Qwen commented on Cline’s announcement that Qwen3.8-Max is available, suggesting users install Cline globally via `npm i -g cline`.

#15 𝕏

NVIDIA AI shared a guide from @MiaAI_lab on chaining DGX Sparks together to run recently released models. The specific models and number of DGX Sparks were not identified.

#16 𝕏

clem 🤗 commented that the new AI model framework treats APIs from providers such as Anthropic and OpenAI differently from open weights, calling the distinction “very good policy.” The thread discusses regulating model weights, APIs, and applications while praising the framework’s treatment of open models.

#17 𝕏

Madhu Guru commented that AI diffusion has been slow because products ask users to navigate technical jargon and complex choices instead of simply completing tasks. Guru predicted a breakthrough product addressing this issue will emerge within the next 12 months.

#18 in

🥞 Carl Vellotti recapped how, after saying no human had read his PRDs since March, he reformatted them with instructions for AI models—including what to quote and “What to tell your human”—and prompted Dave’s AI to call them “unusually well-reasoned.” He says he began specifying which risks to soften in summaries last month, adoption of risky features rose 3x, and discussions about two more engineers began in October before he received two more engineers.

#19 ▶️

FREE Unlimited AI Voices | Better Than ElevenLabs (Microsoft Banned It)

Helena Liu

#20 ▶️

These AI Marketing Agents Get You Customers

Greg Isenberg