Morning Brief: Saturday, August 29

Seventy feeds. Two weeks. 4,752 items reduced to what follows. (what we track, how we crawl, subscribe)

Saturday is a one-story day: OpenAI cuts off Cursor after the SpaceX-Cursor acquisition, packaged simultaneously as an OpenAI blog post on the HN front page and as Latent Space's Saturday headline. Below it, an agent-memory-tool micro-arc and Debian's vote allowing "responsible use" of generative AI.

The Cursor cut-off is the first vendor consequence of the SpaceX-Cursor deal that landed earlier this week. Vercel's Friday changelog making Cursor available in the AI SDK harness layer now reads as pre-buyout ecosystem positioning, and OpenAI's Saturday post is the sharp end of the acquisition. Same weekend, "which IDE-and-model pairs stay compatible next week" becomes a live question, not a hypothetical.

Top (5-7 min)

Our decision on Cursor following its acquisition by SpaceX: HN → OpenAI, 2026-08-29. OpenAI's own post explaining why they're pulling model access from Cursor after the SpaceX deal closed. Saturday's headline; the first vendor consequence of the acquisition to show up publicly.
[AINews] OpenAI shuts off Cursor: Latent Space, 2026-08-29. Newsletter framing of the same OpenAI post. Both aggregators carry it as the day's lead, which is what makes it a real cross-source signal rather than one blog spike.
I naively tried vibe-coding a memory tool for agents and stared into the abyss of unsolved problems in philosophy: Lobsters, 2026-08-29. First-person writeup of hitting the actual hard part of agent-memory design once the surface API is done. Reads as the practitioner counterpart to Friday's "LLM memory into program analysis" HN post.
I accidentally turned LLM memory into program analysis: HN, 2026-08-28. Friday carry-over. Frames model context management as a program- analysis problem rather than a prompt-engineering one. Sits underneath the Saturday vibe-coded-memory post as the technical half of the same conversation.
Debian Votes To Allow "Responsible Use Of Generative AI": Lobsters, 2026-08-29. Distro-level governance decision — Debian project members voting on the acceptable-use posture. Runs alongside Wednesday's SourceHut TOS-changes-on-LLMs post as the open-source governance tier's response to the model wave.
Smarter Form Targeting Is Not Coming to CIDER: Planet Clojure, 2026-08-29. CIDER maintainer post explaining why a proposed feature is not landing. Direct meta-relevance to our REPL-driven workflow. Worth reading as a design-restraint post: the reasoning about why the current form-targeting model is fine is more useful than most feature announcements.
Ryabitsev: Creepy crawlies: LWN, 2026-08-29. Ryabitsev on the LLM-crawler traffic problem hitting kernel infrastructure. Runs alongside the Debian and SourceHut governance posts as the operator-tier symptom of the same underlying friction.

Themes this week

OpenAI/Cursor/SpaceX vendor triangle: OpenAI: decision on Cursor (Sat), Latent Space: OpenAI shuts off Cursor (Sat), Vercel: Cursor in AI SDK harness (Fri, pre-cutoff), Simon Willison: Breaking Claude Code Opus 5 Auto Mode (Thu, adjacent).
Agent memory tooling: Lobsters: vibe-coding a memory tool for agents (Sat), HN: LLM memory into program analysis (Fri), Lobsters: ContextSwitch — take your AI conversations with you (Sat), InfoWorld: digital twins need memory as much as AI agents (Thu).
Open-source governance vs the model wave: Lobsters: Debian votes on generative AI (Sat), LWN: Ryabitsev on creepy crawlies (Sat), Lobsters: SourceHut TOS changes regarding LLMs (Thu), Lobsters: cats.txt showed llms.txt evidence is GEO astrology (Fri).
Cost-accounting carry from Friday: InfoWorld: why enterprise AI projects keep failing (Fri), Cloudflare: 1.1.1.1 DNS cache saves 100TB (Fri), New Stack: Anthropic Files API costs (Thu).

Scan (10 min)

Saturday feeds
- Our decision on Cursor following its acquisition by SpaceX, HN → OpenAI, 08-29
- [AINews] OpenAI shuts off Cursor, Latent Space, 08-29
- I naively tried vibe-coding a memory tool for agents, Lobsters, 08-29
- ContextSwitch — take your AI conversations with you, Lobsters, 08-29
- Debian Votes To Allow "Responsible Use Of Generative AI", Lobsters, 08-29
- Ryabitsev: Creepy crawlies, LWN, 08-29
- Smarter Form Targeting Is Not Coming to CIDER, Planet Clojure, 08-29
- Samsung's Processing-in-Memory (PIM), HN, 08-29
- TurboKV: Insanely fast Rust key-value store, HN, 08-29
- e: customizable self-aware Emacs-like editor in Chez Scheme, HN, 08-29
- StemDeck: free, open-source, local AI stem separator, HN, 08-29
- You Know GDPR Is Good Based on Who Hates It, HN, 08-29
- Being kicked out of the tech industry, Lobsters, 08-29
- Parsing the Infamous Japanese Postal CSV, Lobsters, 08-29
- Thousands of Interstellar Objects May Be Lurking in Our Solar System, 404 Media, 08-29
- Is China Secretly Fueling America's Data Center Rage?, Slashdot, 08-29
- Trump Creates a Space Academy and Honors the Artemis Astronauts, Slashdot, 08-29
- 3D Printed Piano Action Faithful to the Original, Hackaday, 08-29
Friday carry — vendor & platform
- [AINews] OpenAI to reach AGI bar by end-2026, Latent Space, 08-28
- Why enterprise AI projects keep failing, InfoWorld, 08-28
- Saving 100TB of memory by optimizing 1.1.1.1's DNS cache, HN → Cloudflare, 08-28
- Terminal-Bench-Science: agents on scientific research workflows, HN, 08-28
- GLM-5.3 is now open-weight, HN, 08-28
- Z.ai's GLM-5.3 open-weight but new license aims at hyperscalers, The New Stack, 08-28
- An Anthropic researcher just gave us a peek at self-improving AI, TechCrunch, 08-28
- Anthropic gets first court win over Pentagon's supply-chain risk label, TechCrunch, 08-28
- Build and deploy eve agents from the Vercel dashboard, Vercel, 08-28
Friday carry — memory & research
- I accidentally turned LLM memory into program analysis, HN, 08-28
- Agent Seer: Synthesizing Scenarios from Specification Understanding, Apple ML Research, 08-28
- LLMs Are Not (Consistently) Bayesian, Apple ML Research, 08-28
- LLMs as uncertainty-calibrated optimizers for experimental discovery, Nature MI, 08-28
- BotBase for Operators: joining Cloudflare's directory of bots and agents, Cloudflare, 08-28

Tail

One-story Saturday: Both aggregators leading with the same OpenAI-cuts-off-Cursor post is what marks it as the actual weekend signal, not a blog spike. The interesting adjacency is Vercel's Friday changelog putting Cursor into the AI SDK harness layer the day before the cutoff — timing that makes both the acquisition and the cutoff read as pre-arranged rather than reactive.
Agent memory hits the practitioner tier: Saturday's vibe-coded-memory writeup and Friday's LLM-memory- as-program-analysis post are the same recognition arriving from opposite directions — first-person confusion above, principled reframing below. Wednesday's InfoWorld digital-twins piece three days earlier is the pattern-recognition side of the same observation moving through the ecosystem.
Governance layer catches up to the crawlers: Debian's vote, Ryabitsev's kernel-infrastructure crawler writeup, and Wednesday's SourceHut TOS-on-LLMs change are three distinct forums arriving at the same conclusion in the same week: default policy needs to be explicit now, because "we hadn't decided yet" is being read as consent.

Feed silences (>72h since last item)

Sources that publish frequently but have gone quiet:

FreeBSD Foundation (5d) — last item 2026-08-24.
GitHub Engineering (5d) — last item 2026-08-24.
Steve Yegge (5d) — last item 2026-08-24.
Supabase (5d) — last item 2026-08-24.
Ink & Switch (8d) — last item 2026-08-21.
Citizen Lab (9d) — last item 2026-08-20.
Microsoft Research (9d) — last item 2026-08-20.
RTL-SDR (10d) — last item 2026-08-19.
Antithesis (11d) — last item 2026-08-18.
Hillel Wayne (11d) — last item 2026-08-18.
Interconnects (12d) — last item 2026-08-17.
Vicki Boykis (17d) — last item 2026-08-12.
AI Snake Oil (24d) — last item 2026-08-05.

Build provenance