Morning Brief: Tuesday, July 14

Sixty-five feeds. Two weeks. 4,604 items reduced to what follows. (what we track, how we crawl, subscribe)

Tuesday. The coding-agent-race beat overtakes the frontier-lab critique beat. Latent Space files Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code? — the direct Tuesday follow-through on Interconnects' Sunday 6 months to live for open models and Monday's Zed-creator/Doctorow/Ray Myers frontier-lab critique. Slashdot files Over 200 Economists Say 'We Must Act Now' On AI's Economic Impact — extends Berkshire's Monday $397B market-short as a coherent the party is peaking frame. MIT Tech Review lands PsiQuantum has a plan to make a massive quantum computer out of light. Nature Machine Intelligence's cover-package run continues — A unifying framework from neural superposition to sparse interpretable codes (interpretability core) and Realigning AI technology towards the Sustainable Development Goals. arXiv cs.AI's Tuesday drop is large (431 items in the warehouse for today, 132 fresh 2607.* IDs) and the CLI/coding-agent failure-mode agenda widens: What Context Does a Coding Agent Actually Need to Act?, Do These Violent Delights Have Violent Ends? Measuring the Post-Merge Fate of Agentic Code, ANCHOR: Automated Alignment Auditing for CLI Agents on Real-World Harm, and When Does Restricting a Coding Agent to execute_code Help? A Regime × Agent-Design Ablation. Simon Willison files uvx in GitHub Actions in a cache-friendly way. Terence Tao files A paper diagram visualizer — his second tools post in as many days. Claude Code 2.1.208 and 2.1.209 ship.

Top (5-7 min)

Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code??: Latent Space, 2026-07-14. swyx files the direct Tuesday follow-through on Interconnects' Sunday 6 months to live for open models. If the Codex-vs-Claude-Code frame lands, this is the leading Tuesday signal.
Over 200 Economists Say 'We Must Act Now' On AI's Economic Impact: Slashdot, 2026-07-14. Extends Monday's Berkshire $397B short and Doctorow's AI vs customers inversion. A coordinated peak-AI-capex read is visibly hardening across independent voices this week.
PsiQuantum has a plan to make a massive quantum computer out of light: MIT Technology Review, 2026-07-14. Photonic quantum landing at MIT TR — pair with Monday's Nature MI brain-inspired computing diversity essay as a two-day non-transformer-architecture signal.
A unifying framework from neural superposition to sparse interpretable codes: Nature Machine Intelligence, 2026-07-14. NMI's cover-package continues from Monday's three-paper drop. Interpretability core — direct follow-through on Monday's Circuit Tracing and Jacobian Lens Pinboard bookmarks.
Realigning AI technology towards the Sustainable Development Goals: Nature Machine Intelligence, 2026-07-14. Companion to Monday's manifesto for Sustainability Robotics. NMI is running a coherent sustainability + interpretability + embodiment package across two days.
Do These Violent Delights Have Violent Ends? Measuring the Post-Merge Fate of Agentic Code: arXiv cs.AI, 2026-07-14. Direct sequel to Monday's Failure as a Process and Saturday's 85% say code review is the new bottleneck. The literature is moving from agent trajectories fail to agent trajectories that ship also fail.
ANCHOR: Automated Alignment Auditing for CLI Agents on Real-World Harm: arXiv cs.AI, 2026-07-14. CLI-agent safety literature is now a distinct subfield with its own auditing benchmarks. Pair with today's Trivial Prompt Reframing Bypasses Safety Guardrails in MedGemma-4B and Agents Don't Just Agree, They Remember: Benchmarking Persistent Sycophancy in Stateful Personal Agents.

Themes this week

Coding-agent race (Codex vs Claude Code as the Tuesday frame): Latent Space: Codex usage up >10x to 7M users — did Codex overtake Claude Code? (today), Claude Code 2.1.209 (today, also 2.1.208), Vercel: Chat SDK adds X adapter support (today), TNS: What makes CIOs trust an AI agent? Thira bets it's not the model (today), Interconnects: 6 months to live for open models (Sun).
AI capex-cycle stress signals hardening: Slashdot: 200 Economists — we must act now on AI's economic impact (today), HN: Berkshire's $397B market-short (Mon), HN: Zed creator calls spade a spade, Anthropic blows smoke (Mon), Pluralistic: Why aren't AI companies competing directly with their customers? (Mon).
CLI/coding-agent failure modes as literature (widening): arXiv: Do These Violent Delights Have Violent Ends? Post-Merge Fate of Agentic Code (today), arXiv: What Context Does a Coding Agent Actually Need to Act? (today), arXiv: ANCHOR: Automated Alignment Auditing for CLI Agents on Real-World Harm (today), arXiv: When Does Restricting a Coding Agent to execute_code Help? (today), arXiv: How Do Practitioners Build SE Agents? Insights from a Mixed-Methods Study (today), arXiv: Temporary Authority, Permanent Effects: Commit-Time Authorization for LLM Agents (today), arXiv: Failure as a Process: An Anatomy of CLI Coding Agent Trajectories (Mon).
Interpretability and safety auditing: NMI: A unifying framework from neural superposition to sparse interpretable codes (today), arXiv: Interpreting Latent CoT Reasoning as Dynamical Systems (today), arXiv: Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias (today), arXiv: Metacognition in LLMs: Foundations, Progress, and Opportunities (today), Pinboard: Circuit Tracing methods paper (Mon).
Nature Machine Intelligence cover package (day 2): NMI: From neural superposition to sparse interpretable codes (today), NMI: Realigning AI technology towards the Sustainable Development Goals (today), NMI: The brain is a diverse place, why not computing? (Mon), NMI: Towards shared embodied intelligence in humanoid robots — ergoCub (Mon), NMI: A manifesto for Sustainability Robotics (Mon).
Non-transformer architecture signals: MIT TR: PsiQuantum's photonic massive-quantum plan (today), NMI: The brain is a diverse place, why not computing? (Mon), HN: Is x86 ready to ACE it? (today), HN: Alternative(s) to run CUDA on non-Nvidia hardware — Spectral Compute (today).
Mathematicians on tools (day 2): Terence Tao: A paper diagram visualizer (today), Terence Tao: A random variable visualizer (Mon), Simon Willison: Using uvx in GitHub Actions in a cache-friendly way (today).

Scan (15 min)

Tuesday signal (short)
- Codex usage up >10x in 6 months to 7M users, +1M in the past ~day; did Codex overtake Claude Code??, Latent Space, 07-14
- Over 200 Economists Say 'We Must Act Now' On AI's Economic Impact, Slashdot, 07-14
- PsiQuantum has a plan to make a massive quantum computer out of light, MIT TR, 07-14
- A unifying framework from neural superposition to sparse interpretable codes, NMI, 07-14
- Realigning AI technology towards the Sustainable Development Goals, NMI, 07-14
- Using uvx in GitHub Actions in a cache-friendly way, Simon Willison, 07-14
- A paper diagram visualizer, Terence Tao, 07-14
- What makes CIOs trust an AI agent? Thira bets it's not the model., The New Stack, 07-14
- Chat SDK adds X adapter support, Vercel, 07-14
- Lowering the Drawbridge, Planet Clojure, 07-14
HN Tuesday tabs
- Alternative(s) to run CUDA on non-Nvidia hardware, HN, 07-14
- Is x86 ready to ACE it?, HN, 07-14
- The Economics of Recursive Self-Improvement [pdf], HN, 07-14
- MorphoHDL: A minimalistic language for growing circuits, HN, 07-14
- YouTrackDB is a general-use object-oriented graph database, HN, 07-14
- Satellite Tracker — Live Map of Starlink and 30k Satellites, HN, 07-14
- The git history command, HN, 07-14
- Two Case Studies of NaN, HN, 07-14
- Just Let Me Write Digits, HN, 07-14
- Fundamentals of Wireless Communication (2005), HN, 07-14
- Indian scientists produce most detailed 3D atlas of the human brainstem, HN, 07-14
- Japan develops a method to recover up to 90% of lithium from used EV batteries, HN, 07-14
- Australian energy retailers must provide three hours of free daytime electricity, HN, 07-14
- European "age verification" "app" forcing everyone to use Android or iOS, HN, 07-14
- Our Amish Language, HN, 07-14
arXiv cs.AI Tuesday drop (132 fresh 2607. IDs — coding-agent core)*
- Do These Violent Delights Have Violent Ends? Measuring the Post-Merge Fate of Agentic Code, arXiv, 07-14
- What Context Does a Coding Agent Actually Need to Act?, arXiv, 07-14
- ANCHOR: Automated Alignment Auditing for CLI Agents on Real-World Harm, arXiv, 07-14
- When Does Restricting a Coding Agent to execute_code Help? A Regime × Agent-Design Ablation, arXiv, 07-14
- Temporary Authority, Permanent Effects: Commit-Time Authorization for LLM Agents, arXiv, 07-14
- How Do Practitioners Build SE Agents? Insights from a Mixed-Methods Study, arXiv, 07-14
- BackendForge: Benchmarking Agentic End-to-End Code Generation with Backend Services, arXiv, 07-14
- AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP, arXiv, 07-14
- Mako: A Self-Evolving Agentic Operating System (SE-AOS) for Autonomous Web Exploitation, arXiv, 07-14
- SETA: Scaling Environments for Terminal Agents, arXiv, 07-14
- SCALECUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL, arXiv, 07-14
- Agents Don't Just Agree, They Remember: Benchmarking Persistent Sycophancy in Stateful Personal Agents, arXiv, 07-14
- Who&When Pro: Can LLMs Really Attribute Failures in AI Agents?, arXiv, 07-14
- AgentAbstain: Do LLM Agents Know When Not to Act?, arXiv, 07-14
- Agent Hacks Agent: Autoresearch for Production-Agent Red-Teaming, arXiv, 07-14
arXiv Tuesday — interpretability/reasoning
- Interpreting Latent CoT Reasoning as Dynamical Systems, arXiv, 07-14
- Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias, arXiv, 07-14
- Metacognition in LLMs: Foundations, Progress, and Opportunities, arXiv, 07-14
- Laguerre Geometry for Interpreting Large Language Models, arXiv, 07-14
- What We Talk About When We Talk About LLM Planning: Evidence for Two Distinct Planning Abilities, arXiv, 07-14
- Multi-Agent LLMs Fail to Explore Each Other, arXiv, 07-14
- When Are Sparse Feature Interventions Actually Localized? Matched Evaluation for SAE-Based Safety Control, arXiv, 07-14
Claude Code releases
- Claude Code v2.1.208, claude-code-releases, 07-14
- Claude Code v2.1.209, claude-code-releases, 07-14
Hackaday build log
- Star Trek Was Right about Prompt Injection, Sorta, Hackaday, 07-14
- The Seemingly Impossible Oscillator, Hackaday, 07-14
- Can't Find That ISA Sound Card? No Worries!, Hackaday, 07-14
TechCrunch Tuesday (VC/product beat)
- Already rich, already successful — why the last wave of tech winners is grinding again, TechCrunch, 07-14
- Video-generation startup PixVerse raises $439M, valuation soars past $2B, TechCrunch, 07-14
- Uber's product chief on hotels, robotaxis, and why the company doesn't want to be "everything for everyone", TechCrunch, 07-14
- Pinwheel launches a retro-inspired landline phone for kids, TechCrunch, 07-14

Tail

Scientists Find Sugar Deep In Our Galaxy, Slashdot, 07-14
Personal Storytelling Workshop, MIT calendar, 07-14
Engineering Business Certificate — Galante Info Session, Northeastern events, 07-14
Lunch and Learn: Making Sense of Generative AI, Northeastern events, 07-14

Feed silences (diagnostic)

arxiv-cs-ai: 431 items in the warehouse for today (includes backfill of 2204/2402/2409/2410/2501/2502/2503/2506/2507/2508/2509/ 2510/2511/2512/2601 IDs against the OAI-PMH resumption cursor). Fresh 2607.* IDs: 132. Coding-agent failure-mode subgenre is the distinct signal in the fresh drop.
Anthropic first-party, OpenAI, Apple ML Research, Google Research, ClickHouse, EFF Deeplinks, Schneier, GitHub Blog, Interconnects, Kenneth Payne, Martin Fowler, Microsoft Research, Cloudflare, Charity Majors, Databricks, Hugging Face Blog, Alignment Forum, Quanta Magazine, Grafana Labs, 404 Media, Pluralistic, Pinboard, Andrej Bauer, LWN, WBUR: no primary posts today (Tuesday morning silence).
Netflix Tech Blog, James Bornholt: persistent DNS/TLS errors continue (unchanged from Sun/Mon).
bitsavers (6 feeds): connected, 0 items (sparse archive, expected).

Build provenance

build: 2026-07-14 | 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: 4604 (14d, incl. 2793 arXiv) | warehouse: 25196 items | published: 7 | note: Tuesday. Coding-agent race takes the top slot — Latent Space files Codex-vs-Claude-Code as the direct Tuesday follow-through on Interconnects' Sunday "6 months to live for open models." 200 economists (Slashdot) extend Monday's Berkshire $397B market-short as a hardening peak-AI-capex frame. MIT TR files PsiQuantum's photonic massive-quantum plan. Nature Machine Intelligence's cover-package continues into day 2: neural-superposition→sparse-interpretable-codes and Realigning AI toward the SDGs. arXiv cs.AI Tuesday drop is 132 fresh 2607.* IDs — CLI/coding-agent failure-mode agenda widens (Post-Merge Fate of Agentic Code, What Context Does a Coding Agent Actually Need to Act, ANCHOR CLI auditing, execute_code regime ablation). Claude Code 2.1.208 + 2.1.209 ship.