Good morning. It’s Thursday, September 4. You can almost taste the weekend. Finish the week strong! 💪
Today, we’ll be covering
GitHub makes agentic coding first-class with an Agents panel; GPT-5 + model options arrive
Notion API v2025-09-03: multi-source databases land (devs: action required)
xAI releases a fast, low-latency coding model for agents
Perplexity Comet Plus bets on publisher revenue sharing
Other News: NVIDIA releases latest on-device model, IBM/NASA launch a solar weather model
Research/dev releases: OpenAI GPT-OSS open-weight models; Mistral Medium 3.1
GitHub turns agentic up to 11
GitHub launched an Agents panel so you can delegate tasks to the Copilot coding agent from any GitHub page and track progress without switching context. Also rolling out: GPT-5 in Copilot + a Smart Mode, and broader agent capabilities across VS, VS Code, Mobile, and Raycast (changelog). Coverage: The Verge; on X: @tomwarren. (The GitHub Blog, Microsoft Tech Community, The Verge)
Why it matters: background agents who can open PRs + model routing = fewer tab flips, faster merges, clearer audit trails.
Notion ships multi-source databases (API v2025-09-03)
Notion’s latest API version separates “databases” from “data sources,” so one database can aggregate multiple underlying tables. Start with Changes by version and the API changelog; confirmation on X: @NotionAPI. If you query /v1/databases, check the deprecations and new endpoints—Data sources now exist and you’ll want to read them before you ship. (Notion API, X (formerly Twitter))
Action for builders: scan your integrations that write/query DBs; map DB IDs ↔ data source IDs, and update any webhook handlers.
xAI drops grok-code-fast-1
xAI released grok-code-fast-1, a fast, cheaper coding model targeting agentic workflows. First coverage ties it to coding assistants (including Copilot preview access). More reporting: Reuters and IT Pro.
TL;DR — a contender for low-latency code editing, planning, and tooling when GPT-class models are too pricey. (xAI, The GitHub Blog, Reuters, IT Pro)
Perplexity’s Comet Plus models a publisher-friendly approach
Perplexity launched Comet Plus ($5/mo) with an 80% revenue share to participating publishers—another angle on the “AI eats clicks” problem. Good overviews: TechRadar and Windows Central; legal backdrop: Reuters on the News Corp case. (Perplexity AI, TechRadar, Windows Central, Reuters)
Why you care: if you publish or license content, watch how “AI + pay publishers” evolves—and what it does to link-through rates.
Other News
NVIDIA Nemotron-Nano-9B-v2 (open weights) is live on HF—solid for on-device or edge inference. Model card. (Hugging Face)
IBM + NASA “Surya”: open-source heliophysics model on HF to predict solar weather—useful for satellite/energy risk. IBM newsroom. (IBM Newsroom)
Latest AI Researches
OpenAI GPT-OSS — new open-weight models (first open weights since GPT-2), mirrored on HF. OpenAI post. (OpenAI)
Mistral Medium 3.1 — latest multimodal frontier model (Aug ’25). Models overview; also see Voxtral speech models if you’re building voice. (Mistral AI Documentation, Mistral AI)
Prompt of the day
“Create a migration checklist for my Notion integration to API 2025-09-03. Include: data-source ID mapping, affected endpoints, test cases, and a rollback plan. Output as a table with owners & due dates.”
Docs: Changes by version · Deprecated DB query → new data-source query (Data sources). (Notion API)