Braid

Braid

By Lenar Kess · Damra VolNewsTechnology
Download on the App Store

All Braid episodes

  • Mythos Got a Permission List

    Mythos 5 moved from a blocked release to a managed permission list, and the rest of the day filled in the same access fight from different sides: who gets the model, the chips, the serving tricks, and the authority to govern agents once they act.

    • CNBC's Mythos 5 report and TechCrunch's follow-up describe the partial reopening of Anthropic's model to selected companies, agencies, and critical infrastructure operators.
    • Techmeme's supply-chain roundup points from accelerators to packaging, memory, and foundry commitments as the next hard constraint around AI infrastructure.
    • DeepSeek's DSpark paper gives builders a concrete inference-efficiency artifact, with the reported speedup sitting closer to serving cost than model capability theater.
    • The agent attestation paper and the Cedar policy paper show researchers turning agent safety into questions of proof, policy translation, and enforceable boundaries.
    • Techmeme's Meta and California item and The Guardian's youth social-media regulation piece put platform liability beside the model and infrastructure stories.
    29 min
  • The Model Gate Got a Guest List

    Today’s Braid starts with the reported GPT-5.6 staggered rollout, then follows the pressure outward: China’s cheaper models, production-agent architecture, and the first visible places where AI costs are showing up in prices and budgets.

    • Axios on the Trump administration and GPT-5.6 reports that early access may be approved customer by customer, which turns a model launch into a negotiated access problem.
    • Axios on China and the U.S. AI alliance pitch gives the global version of the same pressure: allies can prefer the American story and still choose cheaper, available Chinese models.
    • AI Engineer’s “The Log Is the Agent” talk anchors the builder segment around durable logs, replay, and portability rather than another round of abstract agent-memory talk.
    • The deterministic control-plane paper and the Spec Growth Engine paper show the same pressure in research form: agents need explicit state, permissions, and specs that survive the chat window.
    • CNBC on AI spending discipline, plus Axios on Apple and Microsoft price pressure, turns infrastructure cost into something customers and investors can see.
    • The open-source defense letter adds a software-governance note: the stack still rests on maintainers who are starting to organize around shared legal and institutional exposure.
    22 min
  • Memory Became a Financing Problem

    Today’s episode starts with SK Hynix seeking nearly $29.4 billion for AI investment, then moves into the papers trying to make agents testable, governable, and safer in domains where mistakes leave the chat box.

    • CNBC on SK Hynix gives the day’s concrete infrastructure signal: memory demand is turning into capital-market machinery, not only chip roadmaps.
    • RIFT-Bench tests agentic systems through discovered structure and adaptive probes, which makes security evaluation less dependent on one framework.
    • the offensive-security agent analysis shows how agents used for security work can become targets themselves, including secrets exfiltration and sandbox escape paths.
    • Metis separates text memory from code memory and argues that agents need both, depending on cost, reuse, and transfer.
    • RaDaR reports a rare-disease physician-assistance trial, while the medical safety papers show why deployment claims need trials, logs, and domain-specific failure tests.
    24 min
  • When Access Became Part of the System

    Today's Braid follows a pressure test: once AI systems enter sensitive work, the vendor relationship, the permission model, the data path, and the audit trail become part of the technology itself.

    • Techmeme's pointer to the Mythos 5 NSA reporting anchors the lead: parts of the NSA reportedly red-teamed Anthropic's model for classified-system cybersecurity work, then lost access during the Anthropic dispute.
    • The Financial Times report on banned Nvidia chips in China turns export control from policy language into prices, scarcity, and workarounds.
    • The Linux Foundation's Agent Name Service announcement gives the agent segment a concrete artifact: DNS-backed naming, verification, and discovery for internet-facing agents.
    • SAFARI, LemonHarness, and OpenThoughts-Agent show the research side of the same work: agents need fault attribution, bounded workspaces, time awareness, and better training data.
    • The Guardian's Schneier and Sanders essay and its Meta employee-tracking report move accountability into two places builders can inspect: AI output liability and worker-data consent.
    • The open-source coding-agent census and the GUI-versus-CLI benchmark close on measurement: agent adoption and agent execution are now being studied through traces, interfaces, and verifiers instead of vibes.
    25 min
  • When Chip Access Became Diplomacy

    Today’s episode follows AI supply chains as they move from vendor strategy into state coordination, then turns through Google’s pressure points, Oracle’s AI-linked cuts, builder-facing GLM infrastructure, supercomputing concentration, content rights, and security.

    • Techmeme’s Pax Silica report gives the lead: the Netherlands joined a US-led chip supply-chain effort with South Korea and Japan, while Taiwan endorsed it without becoming a signatory.
    • IEEE Spectrum on Europe’s tech sovereignty package sets the counterpoint: Europe is still trying to reduce dependency in the same stack everyone wants to coordinate.
    • Axios on Google DeepMind departures, CNBC on Google’s search position, and Axios on People Inc.’s crawler complaint make Google’s AI-era pressure concrete without treating it as a collapse story.
    • CNBC on Oracle’s 21,000 role reductions and TechCrunch’s 2026 layoff tracker keep the labor conversation tied to a reported number rather than a slogan.
    • Prime Intellect’s prime-rl v0.6.0 post and Perplexity Developers on GLM-5.2 in the Agent API move the model-release item into builder territory: training infrastructure and agent availability.
    • NVIDIA’s TOP500 post and Techmeme’s LineShine report show two infrastructure claims coexisting: national leadership in a fastest-machine ranking and vendor concentration across installed systems.
    • OpenAI’s DayBreak post, Al Jazeera on the Five Eyes warning, and IEEE Spectrum on vibecoding malware close the episode with the security proof problem: offense, defense, and generated code are all becoming more inspectable and more dangerous.
    31 min
  • When Compute Signed a Utility Contract

    Today’s episode follows AI infrastructure into its physical contracts: gas-fired power in West Texas, water claims from Nvidia, local backlash against data centers, export controls, enterprise rollouts, and the strange new place where coding agents meet private equity diligence.

    • The Wall Street Journal on Chevron and Microsoft reported a 20-year gas-fired power agreement for a proposed West Texas AI data center, which turns compute capacity into a named energy project.
    • Axios on Nvidia’s water claim adds the cooling side of the constraint, with Nvidia arguing that next-generation systems can reduce data center water use.
    • Axios on data center backlash and Gallup’s local-opposition polling show that data centers are becoming the public face of AI anxiety.
    • The Associated Press on China’s restrictions explains Beijing’s new export-control and procurement moves against U.S. defense-linked firms.
    • OpenAI’s Samsung announcement and Getty Images’ OpenAI display partnership show OpenAI signing the enterprise and content agreements that make AI products usable inside institutions.
    • The Financial Times on Bain’s AI replicas points to a new diligence test for software companies: can a buyer approximate the product quickly enough to question the moat?
    25 min
  • When Trust Needs a Test Bench

    Today starts with Anthropic moving from restriction politics toward negotiated assurance, then follows that pressure into agent engineering, public-sector contracts, infrastructure finance, and synthetic media provenance.

    • Indian Express reports that President Trump no longer views Anthropic as a national-security threat, which turns the recent access fight into a question of technical evidence and political confidence.
    • Anthropic Project Fetch: Phase Two gives the lab a primary artifact for that conversation: a research program meant to show what the model does under controlled pressure.
    • Martin Fowler and Bayer describe reliable agentic systems as data, state, and review problems, which is a better builder lens than treating agent loops as better prompting alone.
    • Techmeme on Palantir and the NHS, PublicTechnology on PoliceAI, and Reuters on Google in Germany show public institutions turning AI adoption into contract, policing, and liability disputes.
    • Techmeme on Jane Street and CoreWeave and Noema Atlas on LocalLLaMA give two versions of the infrastructure story: capital buying compute and builders trying to loosen centralized model distribution.
    21 min
  • The Million-Token Bill Arrives

    DeepSeek previewed a V4 model family with one-million-token context claims on the same morning that corporate AI spending discipline, coding-agent evaluation papers, and robotics ownership news pointed at a shared practical constraint: intelligence is getting measured by the loop around it.

    • DeepSeek-V4 preview paper claims one million tokens for both Pro and Flash, with lower long-context inference FLOPs and key-value cache use than DeepSeek-V3.2; the open question is how much retrieval and agent memory this displaces in practice.
    • Financial Times on companies reining in AI usage supplies the budget counterweight: once finance teams meter model usage, pilots become governed systems.
    • The grite paper, StaminaBench, Predictive Validity for LLM agents, and AutoPass turn agent evaluation away from single-task demos and toward coordination, stamina, out-of-sample rank stability, and runtime evidence.
    • Startup Fortune on Hyundai and Boston Dynamics reports Hyundai buying SoftBank's remaining 9.65 percent stake for $325 million, which makes the robotics question less about demos than factory ownership, reliability, and patient deployment.
    • ENPIRE and Human Universal Grasping give the technical robotics companion: real-world policy self-improvement, reusable reset and verification loops, and human grasp data at million-frame scale.
    24 min
  • When Compute Needs Permission

    Today’s episode starts with a grid order, because the AI capacity story is becoming less about a single chip purchase and more about permission, scheduling, power, and who gets to coordinate the stack.

    • FERC’s large-load orders give six regional grid operators 60 days to justify or change tariffs for data centers and other large energy users, which turns AI buildout into a market-design question.
    • NVIDIA’s writeup treats flexible AI factories as grid assets, an argument to examine because NVIDIA is an interested infrastructure actor.
    • Axios on Bernie Sanders’ AI plan, plus reactions from Miles Brundage and Garry Tan, shows the AI upside fight moving from tax rates into ownership and exit mechanics.
    • Zixuan Li’s GLM-5.2 app-development results and the GLM-5.2 model page make the open-weight story more concrete: app-building tasks, a one million token context window, and a model people can run outside one vendor account.
    • ToolPrivBench, the Sovereign Execution Broker paper, and NRT-Bench turn agent safety into an authority problem: which tool is selected, who holds credentials, and whether the evaluation measures harm in the environment instead of judging text.
    • the humanoid robotics data standards paper and Humanoid Everyday make the same infrastructure point in the physical world: robots need reusable records of body, task, scene, timing, and outcome before demos become shared capability.
    20 min
  • Who Gets to Approve the Model

    Today’s episode follows a practical tension in AI: the public story says less regulation, while the daily operating reality is access reviews, model blocks, cloud dependencies, water constraints, and private data moving through agents.

    • Axios on the White House AI power center maps the officials now shaping AI decisions, from Howard Lutnick to Scott Bessent and Ryan Baasch.
    • Axios on Trump’s shadow AI policy explains why an anti-regulation posture can still produce case-by-case intervention.
    • Techmeme’s Wired summary on Anthropic and SK Telecom shows how national-security pressure turns into a specific model-access decision.
    • Axios on Arizona data centers grounds the compute story in power, water, grid costs, and local utility planning.
    • Rest of World on Chile’s undersea cable fight shows network routes becoming geopolitical infrastructure, not neutral background.
    • SafeClawBench, TRAP, and the agent-memory paper turn the research block into one builder question: what state, evidence, and private data should survive contact with tools?
    23 min