Executive Summary
Defense Secretary Pete Hegseth has given Anthropic until the end of the week to grant the U.S. military unrestricted access to Claude, escalating a standoff that is now entangled with reports that the Pentagon is also exploring Elon Musk’s Grok for classified systems. The pressure campaign comes as President Trump, heading into a third round of indirect nuclear talks with Iran on Thursday in Geneva, publicly dismissed accounts that the chairman of the Joint Chiefs had warned against war, even as the administration ordered nonessential staff out of the U.S. Embassy in Beirut and kept a “strike-ready” posture in the region. In Europe, Ukraine marked four years since Russia’s full-scale invasion under fresh drone and missile attacks on its energy grid, while Moscow lobbed new nuclear accusations at Britain and France and Washington warned that China’s arsenal is swelling toward a much larger force. Markets, still digesting a roughly $40 billion drop in IBM’s value after Anthropic promoted an AI tool for COBOL modernization, faced another policy jolt after the White House moved to a new 10 percent tariff regime following a Supreme Court rebuke.
AI & Technology
Hegseth Gives Anthropic Days to Lift Claude’s Military “Red Lines”
Defense Secretary Pete Hegseth delivered an ultimatum to Anthropic on Tuesday: grant the Pentagon unrestricted access to its Claude model by the end of the week or risk the cancellation of a $200 million contract and possible designation as a “supply chain risk,” according to people familiar with the discussions. The threat, which also included the possibility of invoking the Defense Production Act to compel cooperation, sharpened a conflict between Washington’s demand for “all lawful” military uses of frontier A.I. and Anthropic’s insistence that its systems not be used for fully autonomous weapons or mass surveillance.
The confrontation is unusually consequential because Anthropic has already been integrated into classified Defense Department networks under a contract awarded last summer, making it less a hypothetical dispute than a fight over operating terms midstream. Pentagon officials, including the department’s chief technology officer, Emil Michael, have argued that lawful military authority should not be constrained by private corporate policy. In that view, a contractor that can veto categories of missions becomes a national security vulnerability, particularly as A.I. tools move from back-office analysis into targeting, electronic warfare and operational planning.
Anthropic has portrayed its position as both an ethical obligation and a safety requirement, warning that relaxing certain restrictions could accelerate a race toward autonomy with unclear command-and-control safeguards. The company has not publicly committed to meeting the Pentagon’s deadline, and it remains unclear how far it is willing to bend without undermining the safety-focused identity that has helped it attract customers and capital.
The administration’s willingness to float the Defense Production Act—traditionally used to direct industrial capacity in wartime or emergencies—has unnerved some investors and executives who see it as a signal that frontier A.I. may be treated less like software and more like strategic infrastructure. If the government compels access in this case, other A.I. companies may face pressure to rewrite “acceptable use” policies that have functioned as de facto governance in the absence of comprehensive federal law.
Pentagon Weighs Grok for Classified Work as Safety Questions Multiply
As the dispute with Anthropic intensified, the Defense Department has also held discussions about integrating Elon Musk’s Grok into classified systems, according to reports from multiple outlets. The talks suggest the Pentagon is trying to widen its options and reduce reliance on any single vendor at a moment when private companies are increasingly willing to impose conditions on how their models can be used.
The interest in Grok comes alongside a growing public record of both ambition and anxiety around military A.I. In one recent test, U.S. Air Force pilots used A.I. aboard an experimental jet to evade a simulated missile, a vignette that officials have pointed to as evidence that machine assistance is moving rapidly from decision-support into tactics. Yet the same week brought new reminders of the costs of failure: Canadian authorities summoned OpenAI representatives after the company did not notify police about a mass shooter’s account, according to Canadian officials, fueling calls for clearer duties to warn and more robust monitoring.
Anthropic, for its part, has faced scrutiny of its own after revising its Responsible Scaling Policy and dropping a pledge that it would guarantee certain safety measures before training new models. Critics say the change reads like a retreat under competitive pressure; the company has argued that its approach remains rigorous and iterative, and that binding commitments can be difficult to define as capabilities evolve. The juxtaposition—Pentagon demands for broader use, alongside evidence of real-world misuse—has left policymakers trying to reconcile speed with control.
It is unclear how quickly any Grok arrangement could move from exploratory talks to deployed systems, particularly given the security certifications and testing required for classified environments. But the message to the market is already being absorbed: Washington appears willing to reward suppliers who accept broad “lawful use” terms, even as allied governments and domestic regulators push in the opposite direction, toward higher accountability.
IBM’s $40 Billion Wipeout Exposes A.I.’s Power to Reprice Legacy Tech
Shares of International Business Machines fell 13.2 percent on Monday, Feb. 23, the company’s steepest one-day drop in more than 25 years, after Anthropic announced “Claude Code,” a tool it said could automate large parts of COBOL modernization. The selloff erased roughly $40 billion in market value and spilled into other IT and services names as investors reassessed the durability of consultant-heavy modernization work that has long generated dependable cash flows.
COBOL, despite decades of predictions of its demise, remains embedded in core financial and government systems. Industry estimates often cited by technologists suggest it underpins about 95 percent of U.S. ATM transactions and 80 percent of in-person credit card swipes, with hundreds of billions of lines still in use worldwide, much of it running on IBM mainframes. Anthropic’s claim—that multi-year modernization projects can be compressed into “quarters”—was received on trading desks as less a product announcement than a challenge to a business model.
IBM has pushed back, arguing that translating COBOL code is only the beginning of modernization and often the easiest portion. Rob Thomas, a senior vice president at IBM, said the hard problems involve redesigning data architecture, replacing runtimes without breaking transaction integrity, and maintaining performance at scale—areas where IBM believes it retains an advantage and where generative A.I. may still require deep domain expertise and careful verification.
Some analysts have suggested the market reaction carried the hallmarks of narrative panic—a “scary bedtime story” about instant disruption—rather than a sober estimate of near-term revenue loss. Even if A.I. speeds code conversion, organizations that run mission-critical systems often move slowly because the cost of failure is high and regulation is strict. Still, the magnitude of the drop illustrated how quickly investors are willing to reprice incumbents when a credible tool threatens to unbundle labor-intensive work into software.
The episode also put pressure on the entire IT services ecosystem, including global outsourcers, which have relied on modernization backlogs. If A.I. reduces billable hours faster than it increases project volume, margins could compress; if it accelerates demand, vendors may pivot to verification, governance and architecture work. The next signal for markets may come less from Anthropic than from how large customers revise modernization budgets and how IBM frames mainframe growth in coming quarters.
Anthropic Accuses Chinese Rivals of “Industrial-Scale” Distillation Attacks
Anthropic accused three Chinese A.I. companies—DeepSeek, Moonshot AI and MiniMax—of running “industrial-scale distillation attack” campaigns to extract capabilities from Claude, alleging the firms created roughly 24,000 fraudulent accounts and generated more than 16 million exchanges with its model. The company said the activity evaded commercial restrictions and geographic blocks, and that it is deploying countermeasures including behavioral fingerprinting and intelligence-sharing with other firms.
Distillation, a common technique in machine learning, can be legitimate when used on a company’s own models; the dispute centers on whether rivals are effectively copying proprietary capabilities by harvesting outputs at scale. OpenAI has previously flagged DeepSeek for similar behavior, and Google has reported an increase in “distillation attacks,” suggesting the tactic is becoming a routine form of industrial espionage in A.I. competition.
Anthropic warned that models produced through illicit distillation could be deployed without comparable safeguards, raising risks of misuse, including in sensitive domains like biological research. Dario Amodei, Anthropic’s chief executive, has urged stricter U.S. export controls on advanced chips, arguing that controlling compute can slow both direct training and the ability to run large-scale extraction campaigns.
The allegations also landed amid a messy debate over who gets to claim the moral high ground on training data. Mr. Musk, whose xAI produces Grok, mocked Anthropic’s complaint on social media and alluded to past litigation over copyrighted material used in training, deriding Anthropic’s models as “misanthropic and evil” in one post. Anthropic has not accepted the comparison, and it is unclear what evidence, if any, Chinese firms will provide to rebut the specific account-creation numbers.
What happens next may hinge less on court filings than on whether the U.S. government treats model extraction as a national security problem akin to chip smuggling. Even if Washington tightens controls, a technical arms race is already underway, with companies investing in watermarking, anomaly detection and access throttling. The broader question—whether frontier A.I. can be protected like a trade secret in a world of cheap inference and mass automation—remains unanswered.
Geopolitics & Security
Trump Denies General’s Iran Caution as Geneva Talks Loom
President Trump on Tuesday rejected reports that Gen. Daniel Caine, the chairman of the Joint Chiefs of Staff, had warned senior officials against launching a military campaign against Iran. Posting on social media, Mr. Trump said the general believed any operation would be “easily won” and would be “leading the pack” if ordered, an emphatic rebuttal to accounts in The Wall Street Journal and Axios that described Caine as urging caution over casualties, air-defense depletion and overstretched forces.
The public denial came as the United States and Iran prepared for a third round of indirect nuclear negotiations on Thursday in Geneva, mediated by Oman. Mr. Trump has said he prefers a deal but has paired diplomacy with threats, warning Iran of a “very bad day” if an agreement is not reached and previously setting a 10-to-15-day window for progress. Iran has rejected demands for zero enrichment and insists its nuclear program is peaceful, a position that leaves negotiators searching for a formula that limits capability without requiring Tehran’s full capitulation.
The military backdrop has grown heavier. U.S. officials have ordered nonessential staff to depart the American embassy in Beirut, and the administration has deployed two aircraft carrier strike groups and additional bombers to the region, a posture one former NATO commander likened to the buildup ahead of the 1990 Gulf War. Iranian officials, warning that U.S. bases would be targeted in retaliation, have signaled they are dispersing decision-making and fortifying nuclear sites in anticipation of a strike.
Inside the administration, the decision-making structure has drawn scrutiny. Reports say Jared Kushner and Steve Witkoff, prominent figures in Mr. Trump’s orbit, are influencing deliberations over whether to pursue limited airstrikes as coercion or contemplate broader attacks aimed at regime change. The White House has projected confidence that Iran will absorb pressure as it did after earlier U.S. actions, but former officials argue conditions inside Iran—economic strain, political brittleness and heightened regional tension—could make its response less predictable.
With talks days away, both sides are signaling resolve, and neither has offered the kind of public concession that would make a diplomatic landing easier. If Geneva produces only procedural progress, the administration may face a choice between extending its deadline or escalating—two paths with very different risks, and neither one easily reversible once set in motion.
Russia Pounds Ukraine’s Power Grid as War Enters Fifth Year
Russia intensified its campaign against Ukraine’s energy infrastructure as the war moved into its fifth year this week, extending a winter strategy Ukrainian officials describe as turning cold into a weapon. Overnight into Tuesday, Russia launched 133 drones and an Iskander-M ballistic missile, according to Ukraine’s air force, which said it shot down 111 drones; 19 strikes still hit 16 locations, leaving damage that included residential buildings and a gas pipeline in the Kherson region.
The broader toll remains grim and contested. The United Nations Human Rights Monitoring Mission in Ukraine has confirmed at least 15,172 civilian deaths and more than 41,000 injuries since the full-scale invasion began on Feb. 24, 2022, cautioning that true figures are likely higher because verification is difficult in occupied and front-line areas. Ukraine, which says Russia now controls about 20 percent of its territory including Crimea and parts of the Donbas seized before 2022, marked the four-year anniversary on Tuesday with visits from European leaders, including Finland’s president, Alexander Stubb, and Sweden’s prime minister, Ulf Kristersson.
Fighting continued across multiple regions. In Dnipropetrovsk, a Russian attack on Pokrovske killed two people and injured five, Ukrainian officials said. In Zaporizhzhia, authorities reported one dead and six wounded from 918 Russian attacks on 36 settlements, while Russian news agencies reported casualties in Belgorod and Bryansk from Ukrainian shelling. The drone war has become its own theater, with Ukraine deploying long-range systems such as the AN-196 Liutyi and Russia saturating air defenses with waves of cheap drones.
The anniversary brought not only commemoration but also a measure of strategic ambiguity. Kyiv’s supporters have pledged solidarity, yet many European governments are navigating domestic politics and budget constraints, while Washington’s stance has grown more transactional in tone. Moscow, through its spokesman Dmitry Peskov, acknowledged that Russia’s objectives “haven’t been fully achieved yet,” language that suggested persistence rather than compromise.
For civilians, the immediate question is whether the grid can be kept running through the remainder of winter and repaired quickly enough to blunt future strikes. For diplomats, the challenge is that the war’s endurance has made maximalist aims harder to abandon: Ukraine seeks security guarantees that match the scale of the invasion, and Russia seeks leverage that battlefield gains alone have not fully delivered.
Moscow Alleges Western “Dirty Bomb” Plot; U.S. Points to China’s Arsenal
Russia’s envoy to the United Nations, Vassily Nebenzia, and the country’s foreign intelligence service, the SVR, accused Britain and France this week of plotting to help Ukraine develop nuclear weapons, including a “dirty bomb,” an allegation Paris and London dismissed as “blatant disinformation.” Moscow offered no documentary evidence in public, but framed the claim as a violation of the Non-Proliferation Treaty and a threat to global security, rhetoric that analysts often view as aimed at shaping international opinion and justifying escalation.
The accusations carried particular resonance because Ukraine relinquished its inherited Soviet nuclear arsenal in the 1990s under the Budapest Memorandums, assurances Kyiv has repeatedly described as hollow after 2014 and 2022. Ukrainian officials have generally insisted they do not seek nuclear weapons and remain committed to nonproliferation treaties, though the war has revived debates in some quarters about whether the country was left dangerously exposed.
At the same time, the United States turned attention to China’s nuclear buildup. Christopher Yeaw, the assistant secretary of state for arms control and nonproliferation, said China was on track to have enough fissile material for more than 1,000 warheads by 2030, calling the expansion “unprecedented, deliberate, rapid and opaque.” With New START lapsed and no successor agreement in place, Washington has urged Beijing to join arms control talks that would reflect a more multipolar nuclear order.
The two narratives—Russia warning of Western proliferation and the United States warning of China’s growth—are not directly linked, but they share a strategic through-line: nuclear rhetoric is increasingly being used as leverage in conventional conflicts and great-power competition. The risk is that allegations, even when disputed, can harden threat perceptions and narrow diplomatic room, particularly in crises where decision-makers already assume the other side is preparing for the worst.
Mexico Kills “El Mencho” With U.S. Help, Igniting Deadly Retaliation
Mexican forces killed Nemesio Oseguera Cervantes, the leader of the Jalisco New Generation Cartel known as “El Mencho,” on Sunday in an operation in the mountain town of Tapalpa, Jalisco, that drew on U.S. intelligence support, according to officials and people briefed on the raid. The violence that followed was swift and wide: retaliatory attacks across several states left at least 70 people dead, including security forces, suspected cartel members and civilians, and prompted school and business closures in some areas.
The United States’ role was unusually explicit. A newly created U.S. Joint Interagency Task Force, Counter Cartel, based in southern Arizona, provided Mexico with what one report described as a “detailed target package,” and the C.I.A. also offered key support. The breakthrough reportedly came from human intelligence tied to surveillance of a romantic partner who was tracked to a property where Oseguera was meeting with his security detail, allowing Mexican special forces to launch the raid after hours of gunfire.
Oseguera, 59, had been one of America’s most wanted fugitives, with a $15 million bounty, and was widely seen as a central figure in the trafficking of fentanyl and methamphetamine into the United States. He founded the CJNG around 2009 and turned it into a cartel known for military-style tactics and brutal messaging. Mexican officials have cast his death as the most significant blow to organized crime since the recapture of Joaquín “El Chapo” Guzmán nearly a decade ago.
Yet past experience suggests that decapitation can splinter cartels rather than end them, producing smaller factions that may fight more openly for territory and revenue. Mexican authorities say they are restoring order, but it is unclear who can command CJNG’s network now, or whether rival organizations will test the group’s control of trafficking routes and ports. For Washington, the operation also deepens a question that has shadowed decades of counternarcotics policy: whether tighter U.S.-Mexico intelligence integration will reduce violence over time, or simply change who wields it.
New Reports Allege Executions and Civilian Harm From Gaza to Ukraine
An independent investigation by Forensic Architecture and Earshot accused Israeli soldiers of firing more than 900 rounds at clearly marked emergency vehicles in Rafah on March 18, 2025, killing 15 Palestinian aid workers, and alleged that some were executed at close range. The report, based on audio, video and witness testimony, also said Israeli forces later crushed the vehicles and attempted to bury the bodies, a claim that would sharpen scrutiny of Israel’s conduct in Gaza if corroborated by official inquiries.
Israeli officials have not accepted the report’s conclusions, and it is unclear what investigative steps Israeli military authorities have taken in response to the specific allegation of close-range executions. The work arrives in a wider context of contentious claims and counterclaims about targeting procedures, the reliability of battlefield evidence and accountability mechanisms in a war that has produced extraordinary civilian suffering.
In Ukraine, Russian soldiers speaking to the BBC described commanders ordering the execution of fellow troops for refusing orders, including one account in which a commander personally shot four soldiers. Such reports are difficult to independently verify, and Russia has not publicly addressed the specific allegations, but they add to a growing body of testimony about coercion and brutality within units on both sides of the front.
The broader region remains tense. U.S. Air Force F-22 stealth fighters have landed in Israel as part of the U.S. buildup, though their mission has not been publicly detailed, and the Lebanese army reported an Israeli attack near a new observation post and said it ordered its forces to return fire. Together, the reports underscore how quickly tactical events and documentation can become strategic, influencing diplomacy, legal exposure and the calculus of escalation.
Duterte Faces ICC Hearing Over Philippines Drug War Killings
Hearings began at the International Criminal Court in The Hague to determine whether prosecutors have sufficient evidence to proceed to a full trial of former President Rodrigo Duterte over alleged crimes against humanity tied to the Philippines’ “war on drugs.” Human rights groups and ICC investigators have estimated that up to 30,000 people may have been killed between 2016 and 2022, far above the Philippine police’s official figure of about 7,000.
Prosecutors have accused Mr. Duterte of being an “indirect accomplice” in systematic killings, citing at least 78 cases of murder and attempted murder and arguing that he incited police and hired killers to “neutralize” suspects. Mr. Duterte, detained since March of last year, has refused to recognize the court’s jurisdiction and has waived his right to appear at the hearings, citing age and frailty; the court dismissed his filing as “speculative,” according to reports.
The proceedings are being closely watched as a test of whether international institutions can hold former heads of state accountable for domestic campaigns that claim the mantle of public safety. The case is also politically resonant inside the Philippines, where narratives of crime control and human rights abuses remain sharply polarized.
Economy & Markets
Trump Imposes 10% Tariff After Court Rebuke, Floats 15% Increase
President Trump imposed a new 10 percent tariff on a broad range of imported goods after the Supreme Court struck down his earlier sweeping levies, ruling 6-3 that he lacked authority to enact them under the International Emergency Economic Powers Act. The new duties rely on Section 122 of the Trade Act of 1974, a provision typically viewed as temporary—limited to 150 days unless Congress approves an extension—and traditionally tied to balance-of-payments concerns rather than the broader geopolitical aims Mr. Trump has attached to tariffs.
Mr. Trump criticized the court’s decision and said he intended to raise the tariff to 15 percent, the maximum rate under Section 122, though an order formalizing that increase had not been issued. The gap between the announced 10 percent and the threatened 15 percent has injected uncertainty into supply chains and pricing, particularly for importers that had already paid duties under the invalidated regime and are now pressing for refunds.
Trading partners moved quickly to assess their options. China said it was reviewing countermeasures and preparing for new high-level trade talks with Washington, while the European Union and Japan have been weighing possible responses, according to officials and diplomats. Analysts say the court ruling may constrain the White House’s ability to use tariffs to pursue non-trade objectives, such as pressuring third countries over energy purchases from Russia, but the administration’s pivot shows it remains determined to keep tariffs central to its economic program.
The next test may come from Capitol Hill and the courts. If the tariffs begin to bite consumers or disrupt critical inputs, lawmakers could face pressure to rein in extensions, while companies could challenge the administration’s interpretation of Section 122. For now, the tariff shift has restored an old reality of Trump-era economics: policy risk is once again a pricing factor.
U.S.-Indonesia Deal Targets Nickel as Tanker Rates Hit Six-Year High
A new trade agreement between the United States and Indonesia is poised to reshape supply chains for nickel, a crucial input for batteries, challenging China’s dominant position in refining and downstream processing. Details of implementation and enforcement will matter, but the deal signals Washington’s continued push to secure critical minerals through friendlier corridors as it tries to reduce vulnerability in clean-energy supply chains.
The agreement unfolded amid a surge in oil shipping costs. Rates for Very Large Crude Carriers on the Middle East-to-China route have climbed to their highest levels since 2020, driven by a mix of geopolitical risk around Iran, shifting trade flows, and heightened demand as India replaces some Russian barrels with Middle Eastern crude. Russian oil, meanwhile, has been moving in greater volumes to China, with Moscow increasingly relying on larger supertankers and ship-to-ship transfers in the Red Sea to avoid scrutiny elsewhere.
Europe’s industrial core is also feeling the strain of changing trade patterns. Germany, confronting what some economists call a “second China shock,” has seen exports to China fall to their lowest level in a decade even as imports rise, a reversal that reflects China’s rapid ascent in high-tech manufacturing, including electric vehicles. German automakers have watched exports to China plunge by roughly two-thirds since 2022, according to reports, pressuring profits at home and abroad as Chinese brands like BYD expand.
The threads converge in commodity markets that have become more political and more fragile. Nickel supply, tanker capacity, and vehicle competitiveness now turn not only on cost curves but on tariffs, export controls, sanctions enforcement, and the probability of conflict. The resulting volatility has made hedging harder and long-term planning more contingent, even for companies that once treated these flows as stable.
Regional Developments
China Targets Japanese Defense Giants With Dual-Use Export Controls
China’s Commerce Ministry on Tuesday placed 40 Japanese entities on export control lists, barring 20—among them Mitsubishi Heavy Industries and Kawasaki Heavy Industries—from importing dual-use items from China and adding another 20, including Subaru and the Institute of Science Tokyo, to a watchlist requiring export licenses and risk assessments. Beijing framed the move as a national-security measure tied to concerns about “remilitarization,” following months of friction that included remarks by Japan’s prime minister, Sanae Takaichi, suggesting possible intervention if China used force against Taiwan.
Japan condemned the restrictions. Sato Kei, the deputy chief cabinet secretary, called the move “deplorable” and said it would “not [be] tolerated” by Tokyo, language that reflected both economic anxiety and a deeper fear that Beijing is willing to weaponize trade to slow Japan’s defense modernization. The inclusion of research institutions and aerospace-linked entities, including references in reports to organizations like the Japan Aerospace Exploration Agency and the National Defense Academy, suggested a broad view of what counts as militarily relevant.
The immediate impact will depend on how exposed the targeted companies are to Chinese-origin components and materials, and whether alternatives can be sourced quickly at tolerable cost. China’s rules also restrict foreign firms from re-exporting China-origin dual-use items to the listed Japanese entities, a tightening that could complicate supply chains beyond Japan.
The decision lands at a moment when Tokyo is already reassessing its security posture amid doubts about U.S. reliability and questions surrounding the planned drawdown of about 9,000 Marines from Okinawa, a long-scheduled realignment now viewed by some analysts as strategically mistimed. Beijing’s controls may therefore function as both economic pressure and political signal, pressing Japan to weigh the costs of deterrence in industrial terms as well as military ones.
War’s Human Toll Shows in Ukraine’s Birth Rate and South Africa’s Repatriations
South Africa’s president, Cyril Ramaphosa, secured the release of 17 South African nationals who said they were misled into fighting for Russia in Ukraine, believing they were headed to bodyguard training before finding themselves on the front lines. Most are expected to return soon, though two remained in Russia, including one hospitalized, according to reports, underscoring how the war has pulled in foreign nationals through opaque recruiting pipelines.
Ukraine, meanwhile, is facing a demographic collapse that may outlast the battlefield. Its fertility rate fell to an estimated 1.00 births per woman in 2025, with some reports placing it as low as 0.8 to 0.9, far below the replacement level of 2.1. Russia’s fertility rate has also declined, projected at 1.37 by 2025, a reminder that the war’s economic and psychological effects are not confined to the invaded country.
The demographic shift has profound implications for reconstruction, labor markets and long-run military readiness, even if a cease-fire emerges. Yet turning grim statistics into policy has proved difficult in a country focused on survival and repair, and in a region where displacement may become permanent for many families.
From the Timeline
“Build for agents,” not just humans: CLIs, demo-first dev tools, and vibecoding-as-distribution
@karpathy framed “legacy” CLIs as an unexpected advantage in an agentic world: terminals are a native interface for AI agents to stitch workflows, build dashboards, and operate products end-to-end—prompting builders to ship agent-accessible surfaces (CLI/MCP/markdown docs) instead of only polished GUIs. In parallel, @levelsio argued for a hardline, anti-VC maker playbook (ship fast, charge day 1, boring tech, low burn), positioning “building in public” as the marketing loop that compounds. @tobi supplied a concrete example—using “vibecoding” to build bespoke motorsport telemetry tools rather than waiting for official ports—while @dhh amplified the sentiment shift that agents are now turning individual makers into small teams.
“CLIs are super exciting precisely because they are a “legacy” technology, which means AI agents can natively and easily use them…”
— @karpathy
Speed becomes the new benchmark battleground: diffusion LLMs and “score-per-task” framing
@AndrewYNg highlighted diffusion LLMs as a credible alternative to autoregressive models, with “reasoning diffusion” positioned explicitly around inference speed as a product edge—not just a research curiosity. Separately, @paulg boosted a lab claiming near-saturation of ARC-AGI-2 with a stated dollar cost per task, signaling how evaluation talk is drifting from raw accuracy toward an economic lens (performance and unit cost), even when details remain thin in-tweet.
Autonomy and embodied AI: open-sourcing “System 1” robot control as the scaling strategy
@DrJimFan made the case that whole-body humanoid control can be scaled like a data problem—motion tracking/mocap as dense supervision—then trained at massive simulation scale for zero-shot transfer, with an emphasis on releasing code and checkpoints. In a parallel but more consumer-facing autonomy thread, @fchollet relayed Waymo’s multi-city rider expansion, reflecting a timeline mood that autonomy is simultaneously “shipping” (robotaxis rolling out) and “platformizing” (generalizable embodied policies).
Wall Street underwrites the AI substrate: bespoke LLM silicon and Japan’s enterprise AI export push
@karpathy zoomed in on the coming “tsunami” of token demand as a hardware-orchestration problem—SRAM vs DRAM tradeoffs, decode latency under long contexts, and the economic prize of getting throughput/latency/$ right—using MatX’s large raise as a marker that specialized chips are back in fashion. In a different slice of “finance meets AI,” @hardmaru framed Citi’s strategic investment in Sakana AI as proof that enterprise-grade AI agents for finance are becoming an exportable product category from Japan, not just a domestic services story. The connective tissue across both is institutional capital treating AI infrastructure and AI-in-finance as strategic, not experimental.
DeFi’s “walkaway test” vs regulatory risk: UK stablecoin caps and Ethereum’s permissionless vision
@brian_armstrong warned UK rulemaking could blunt competitiveness—especially proposals to cap stablecoin holdings—casting regulators as potential “innovation blockers” right as other jurisdictions accelerate. On the protocol side, @VitalikButerin argued Ethereum’s DeFi agenda should be explicitly permissionless, open-source, privacy/security-first, and resilient to team disappearance (“walkaway test”), while rejecting “onchain finance indiscriminately” in favor of systems that minimize centralized chokepoints. The tension in the timeline is whether consumer-protection instincts (caps, constraints) can coexist with DeFi’s core promise of global, permissionless financial access.
“Agentic commerce” goes mainstream—while Silicon Valley flirts with singularity rhetoric
@patrickc signaled Stripe is formalizing “agentic commerce” as a staged progression (near-term practical levels vs the more speculative end-state often discussed online), positioning payments rails as an enabling layer for autonomous agents that buy/sell/services on users’ behalf. In contrast, @pmarca amplified (with a wink) the “2026 Q1 = singularity” framing attributed to Stripe’s CEO—illustrating how quickly serious commerce-roadmap talk gets remixed into accelerationist narrative. Net sentiment: builders and operators are trying to keep the conversation grounded in adoption stages, even as parts of the investor class reward the most sweeping framing.
“1760 Q1 though”
— @pmarca
Solar’s center of gravity shifts: China’s scale, U.S. interconnection bottlenecks, and the “backyard” problem
@chamath used China’s 2025 solar buildout as a wake-up chart, arguing the U.S. opportunity is now less about new utility farms and more about distributed solar + storage—especially given massive grid-connection queues. @DavidSacks amplified a pro-industrial build stance that acknowledges local opposition but frames “not in my backyard” resistance as a barrier to national competitiveness. Together, the thread reads as an emerging consensus that deployment, permitting, and interconnection—not panel tech—are the binding constraints.
Media narratives, partisanship, and institutional legitimacy: escalation over newspapers, committees, and “who’s standing up”
@elonmusk escalated rhetoric against the New York Times (“utterly disgusting”), reflecting a broader pattern of delegitimizing legacy media while also boosting insurgent politics in the UK ecosystem via Restore Britain. On the other side of the aisle, @ylecun amplified claims about withheld Epstein-related material tied to Trump, mirroring how elite tech voices increasingly launder political salvos through retweets rather than original commentary. Meanwhile, @chamath seized on a viral congressional moment to paint Democrats as out of step on “citizens vs illegal aliens,” underscoring how timeline debate is less about policy detail and more about symbolic legitimacy tests (who stands, who’s “captured,” who’s “lying,” which institutions can be trusted).