Morning Brief: Sunday, July 12

Sixty-five feeds. Two weeks. 4,214 items reduced to what follows. (what we track, how we crawl, subscribe)

Sunday. The infrastructure beat opens the day: Cuba suffers its second nationwide blackout in a week — 10 million people, and no grid-scale fix in sight. UK shops will soon instantly alert police about offenders via facial recognition — a real-world productization of the surveillance stack Schneier warned about on Friday. Meta pulls an Instagram AI feature after backlash (the features people did not ask for pattern is now measurable). OpenAI's response to a saturated enterprise market is go where the enterprise isn't: households and families. Saturday's carryover still warrants attention: The New Stack reports Microsoft joining Google in backing Go for AI agents (OpenAI and Anthropic notably absent from the roster); a survey shows 85% say code review is the new bottleneck after coding agents; and a 24-hour race to the bottom on price across OpenAI, SpaceXAI, and Meta. Two mathematicians write Sunday reads on modern coding agents: Terence Tao on old and new apps via modern coding agents and Andrej Bauer on making AI smarter with AI. On the tooling front: Simon Willison shipped sqlite-utils 4.1, and HN found the Mindwalk demo — replay coding-agent sessions on a 3D map of your codebase, which is one of the more interesting explorations of the Genson3 "how did this agent get here?" problem.

Top (5-7 min)

10 Million Cubans Suffer Nationwide Blackout — For The Second Time This Week: Slashdot, 2026-07-12. Second nationwide blackout in one week. The Cuban grid story is becoming an ongoing infrastructure-collapse case study — worth tracking as a real-world analog for what "graceful degradation" does and does not mean at national scale.
Facial Recognition in UK Shops Will Soon Instantly Alert Police About Offenders: Slashdot, 2026-07-12. Real-world productization of the surveillance stack: private CCTV feeds wired into a police-alert pipeline, with the "offender" label determined upstream. Read against Friday's Schneier piece on AI surveillance and social progress, and Saturday's Flock-cameras story (a journalist wrongly tracked and arrested).
Meta Removes Controversial AI Feature On Instagram After Backlash: Slashdot, 2026-07-12. The backlash-then-retract pattern is now measurable: features shipped, users react, feature yanked. This is the friction the agent-control-plane debate keeps missing — end-users have veto power via public revolt.
OpenAI bets on families as ChatGPT goes deeper into households: TechCrunch, 2026-07-11. Read together with Wednesday's "ChatGPT is now a partner for your most ambitious work" and Friday's Codex-into-ChatGPT superapp move: OpenAI is pursuing every user segment simultaneously. Households are the segment enterprise-focused competitors most under-serve.
Microsoft joins Google in backing Go for AI agents — OpenAI and Anthropic lag: The New Stack, 2026-07-11. Two frontier labs and two hyperscalers now diverge on runtime choice for agent frameworks. The Python-first story that dominated 2024–2025 is genuinely fragmenting on production concerns.
85% say code review is the new bottleneck. Here's what the AI coding narrative leaves out.: The New Stack, 2026-07-11. Survey data: the coding-agent productivity story hits a merge gate. Directly relevant to the Every IC engineer is a front-line manager now piece from Friday — throughput moved, so the review queue is the constraint.
Old and new apps, via modern coding agents: Terence Tao, 2026-07-11. Tao files a Sunday-adjacent read on what modern coding agents let a working mathematician build. Companion to Bauer's Making AI smarter with AI — two mathematicians, two orthogonal takes.

Themes this week

Surveillance productization goes retail: Slashdot: UK shops facial recognition alerts police (today), Slashdot: How Flock Cameras wrongly tracked a journalist (Sat), Schneier: AI Surveillance and Social Progress (Fri), Slashdot: Russia hacks doorbell cameras (Fri).
AI feature backlash and the veto surface: Slashdot: Meta removes controversial AI feature on Instagram (today), TC: OpenAI bets on families (Sat), TNS: Anthropic wants you to use AI to decide whether to use AI (Fri).
Go for agents, and the runtime split: TNS: Microsoft joins Google in backing Go for AI agents (Sat), TNS: OpenAI, Microsoft & Anthropic disagree on what you can take back (Fri), TNS: enterprise AI benchmarks are broken (Fri).
Coding agents at the merge gate: TNS: 85% say code review is the new bottleneck (Sat), TNS: The impressive AI demo is dead (Sat), TNS: Every IC engineer is really a front line manager now (Fri), GitHub Engineering: Better tools made Copilot code review worse (Fri).
Mathematicians on coding agents: Terence Tao: Old and new apps via modern coding agents (Sat), Andrej Bauer: Making AI smarter with AI (Sat), Terence Tao: Gilbreath's conjecture (Sat).
Frontier lab price and product churn: TNS: In 24 hours OpenAI, SpaceXAI, and Meta turned AI into a race to the bottom on price (Sat), Vercel: Seedream 5.0 Pro on AI Gateway (Sat), Latent Space: not much happened today (Sat).
Alignment theory and interpretability: Alignment Forum: the current bottleneck is political will, not research (Sat), Alignment Forum: value generalisation — value correction (Fri), MIT TR: Anthropic found a hidden space where Claude puzzles over concepts (Fri).

Scan (15 min)

Sunday signal (short list — quiet weekend)
- 10 Million Cubans Suffer Nationwide Blackout — Second Time This Week, Slashdot, 07-12
- Facial Recognition in UK Shops Will Soon Instantly Alert Police About Offenders, Slashdot, 07-12
- Meta Removes Controversial AI Feature On Instagram After Backlash, Slashdot, 07-12
- Reed Jobs would rather talk about curing cancer than his last name, TechCrunch, 07-12
- Show HN: Mindwalk — Replay coding-agent sessions on a 3D map of your codebase, HN, 07-12
- A dock that wakes up reliably, HN, 07-12
- Why Write Code in 2026, HN, 07-12
Saturday carryover worth re-reading Sunday morning
- Microsoft joins Google in backing Go for AI agents — OpenAI and Anthropic lag, The New Stack, 07-11
- 85% say code review is the new bottleneck, The New Stack, 07-11
- The impressive AI demo is dead. Here's what actually reaches production, The New Stack, 07-11
- In 24 hours OpenAI, SpaceXAI, and Meta turned AI into a race to the bottom on price, The New Stack, 07-11
- OpenAI bets on families as ChatGPT goes deeper into households, TechCrunch, 07-11
- Old and new apps, via modern coding agents, Terence Tao, 07-11
- Making AI smarter with AI, Andrej Bauer, 07-11
- Gilbreath's conjecture: a Cramér random model and deterministic analysis, Terence Tao, 07-11
- sqlite-utils 4.1, Simon Willison, 07-11
- The current bottleneck is political will, not research, Alignment Forum, 07-11
- Scientists Solve Mystery of Bizarre 'Alien Megastructure' Star, 404 Media, 07-11
arXiv weekend batch (cs.AI, still fresh from Saturday's 216-item drop)
- From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents, arXiv cs.AI, 07-11
- Overthinking: Amplifying Reasoning Weights to Extract Learned Secrets, arXiv cs.AI, 07-11
- Remember When It Matters: Proactive Memory Agent for Long-Horizon Agents, arXiv cs.AI, 07-11
- Context Graphs for Proactive Enterprise Agents, arXiv cs.AI, 07-11
- Game Theory Driven Multi-Agent Framework Mitigates LM Hallucination, arXiv cs.AI, 07-11
- When LLMs Agree, Are They Right? Auditing Self-Consistency as Confidence, arXiv cs.AI, 07-11
- Workflow as Knowledge: Semantic Persistence for LLM-Mediated Workflows, arXiv cs.AI, 07-11
- CausalDS: Benchmarking Causal Reasoning in Data-Science Agents, arXiv cs.AI, 07-11
HN weekend tabs
- Nvidia, CoreWeave, and Nebius: Inside the Circular Financing of the GPU Boom, HN, 07-11
- Mesh LLM: distributed AI computing on iroh, HN, 07-11
- Prefer strict tables in SQLite, HN, 07-11
- Good Tools Are Invisible, HN, 07-10
Hackaday build log
- Porting DOOM to the Casio Loopy, Hackaday, 07-12
- Software-Defined Vehicles Loom Closer Every Year, Hackaday, 07-12
- MicroPython is this Summer's Hottest Title for the SNES, Thanks to Claude Fable, Hackaday, 07-11

Tail

Jacobian Lens — Qwen3.6-27B on Neuronpedia, Pinboard, 07-12
People Keep Sneaking Into an Empty IBM Campus — and Then Getting Arrested, Slashdot, 07-11
How the FSF Sysadmins are Blocking Botnets with reaction, Slashdot, 07-11
GPT-5.6 Sol Ultra produces proof of the Cycle Double Cover Conjecture (unverified), HN, 07-10

Feed silences (diagnostic)

arxiv-cs-ai: no new posts today — expected Sunday cliff, Saturday's 216-item drop is still the active reading batch. Monday will reset.
Anthropic first-party, OpenAI, Simon Willison, Apple ML Research, Google Research, ClickHouse, EFF Deeplinks, LWN, Planet Clojure, Schneier, GitHub Blog, Interconnects, Kenneth Payne, Martin Fowler, Microsoft Research, Cloudflare, Charity Majors, Databricks, Hugging Face Blog, MIT Technology Review, Nature Machine Intelligence, Alignment Forum, Quanta Magazine, Grafana Labs, Vercel, The New Stack: no primary posts today — the expected Sunday silence across first-party publishing channels.
Latent Space: no post today (yesterday's not much happened headline holds).
404 Media: no primary post today.
Netflix Tech Blog, James Bornholt: persistent DNS/TLS errors continue (unchanged).
WBUR: no Endless Thread today.
bitsavers (6 feeds): connected, 0 items (sparse archive, expected).
claude-code-releases: no release today.

Build provenance

build: 2026-07-12 | crawler-sha: 5fe7ab8 (Walsh-Research/1.2, compliance v1.3) | feeds: 65 active (78 configured, incl. 12 corp-eng, 6 bitsavers, 5 generated) | items-considered: 4214 (14d, incl. 2397 arXiv) | warehouse: 24360 items | published: 7 | note: Sunday. Cuba's second nationwide blackout of the week, UK retail facial recognition wired to police alerts, Meta pulls an Instagram AI feature after backlash. OpenAI's household push (families) as saturation strategy. Saturday carryover: Microsoft joins Google backing Go for AI agents (OpenAI/Anthropic lag), 85% call code review the new bottleneck, 24-hour OpenAI/SpaceXAI/Meta price war. Two mathematicians file Sunday reads on coding agents (Tao, Bauer). arXiv weekend still fresh: harness engineering for auditable agents, Overthinking secret-extraction, proactive memory for long-horizon agents.