• LNK Energies pitches a nonflammable, room-temperature liquid hydrogen carrier that can use existing diesel delivery infrastructure; it claims this cuts the delivery-cost component from 95% to 6%, makes hydrogen 26% cheaper than diesel, and has completed 50 hours of lab testing TechCrunch. It is targeting drayage trucks, plans a California commercial truck with Symbio within 18 months, and is raising a seed round; the company says permits are not yet secured and identifies engineering and manufacturability as remaining hurdles TechCrunch.
• Industrial Next says its AI-equipped machines have produced more than 7 million smartphones at about a three-second cycle time, including a station that began producing a previously unseen phone model within 60 seconds; its founding team brings Tesla factory/autopilot and Waymo/Meta robotics experience TechCrunch. It says manufacturing-specific precision and speed led it to build its own foundation model, trained largely in simulation with about two hours of real-world data; its announced BMW partnership is an evaluation of a few tasks, and software licensing is the recurring revenue component TechCrunch.
• Neobia AI is building an equipment-connected AI process engineer using fine-tuned open-source models; it says a defect investigation at an injection-molding plant fell from two weeks to three hours, and its edge offering is designed for factories that keep equipment offline and data on-site TechCrunch. It reports 32 companies in its pipeline, three active pilots, four signed LOIs, $65,000 in logged revenue, and more than $600,000 in “re-qualified” pipeline revenue; it charges 2% of production-line capex and is starting with thin-film coatings, which it sizes at $8 billion TechCrunch. The product is currently at autonomy level 2, where AI detects defects and recommends fixes but human approval is required TechCrunch.
• Helix Carbon proposes containerized electrochemistry that converts CO₂ in dirty blast-furnace waste gas back into carbon monoxide for reuse, aiming to cut coke use without replacing existing furnaces or adding carbon capture and gas separation TechCrunch. From a 0.5-kW bench system, it models about $14 per ton in savings without green premiums or incentives at electricity prices around $0.05/kWh; it says economics weaken around $0.07/kWh and the system can operate intermittently TechCrunch. It announced a 3-kW Minnesota research-furnace pilot for Q1 next year, followed about 12 months later by a 150-kW commercial unit under purchase order with a Fortune 500 company, and acknowledged some scale-up risk TechCrunch.
• The lithium-enrichment startup says it is producing pilot batches of space-grade lithium and aims to supply lithium-6 enriched to nuclear grade for fusion, with a first nuclear-grade shipment targeted around 2029; it reports 50+ LOIs and $100 million in offtake interest TechCrunch. Customers are still moving through sample stages—1 kg, then 100 kg, before metric-ton sales—and the company names sample validation and binding offtake as future milestones TechCrunch. Space-grade sales to aerospace, AI/chip and other markets are its hedge if fusion is delayed TechCrunch. Its funding disclosures differ: the pitch says more than $50 million in equity and grants raised, while the Q&A says about $15 million raised and roughly $50 million needed for first commercial production; the founder identifies manufacturing and operations, rather than science, as the main challenge TechCrunch.

• Longshot Space is developing a ground-based multi-injection launcher with a hypersonic lifting body and solid-rocket second stage; the founder said a demonstrator is operational and a 30-inch-interior-diameter gun is targeted for December. TechCrunch The founder reported about $20 million raised, letters of intent, roughly $15 million in Department of War hypersonic-technology development bookings, and a current Series A raise. TechCrunch Its proposed early payloads are rugged electronics such as semiconductors, not large optical instruments or 20-ton geostationary satellites; the founder says remaining technical challenges center on heat-shield survival and navigation at Mach 14–16, rather than the accelerator. TechCrunch
• Veraritoss embeds secure microchips in products and links them to digital certificates and a brand dashboard; the company says its flexible, washable, waterproof chips include a tamper wire that alerts brands if a chip is removed, unlike copyable QR/NFC solutions. TechCrunch Its iOS app is live, the founder said products are already with consumers, and the next-generation chip scans on metal and through glass; the company announced an Athos partnership to authenticate Tiffany & Co. championship crowns. TechCrunch Veraritoss sells directly to brands and charges for hardware plus a software subscription. TechCrunch
• Llas Intelligence combines generative world models with testing agents to create simulated physical environments and automatically explore edge cases for autonomous vehicles and robotics. TechCrunch The founder contrasted an earlier test-track digital twin that cost $250,000 and 15 person-months with a newer approach costing about $600 and 300 GPU hours. TechCrunch The company reports paid top automotive OEM and autonomous-driving customers, robotics design partners, a US DOT RPI challenge win, and work with NVIDIA Metropolis and San Jose. TechCrunch The founder identifies the sim-to-real gap and broad acceptance of shared safety evidence by insurers and regulators as key hurdles; the pitch also cited an open-source Cosmos world model competitor. TechCrunch
• The wildlife-AI startup, introduced as Flock Intelligence and called Flux Intelligence by its founder, uses motion-triggered computer vision to identify animal species, age, sex, and behavior, then plays context-specific bioacoustic signals on-device without cloud processing. TechCrunch The company reports deployments across three continents, including five US international airports, and two years of daily train deployment with Alstom, with joint commercialization being prepared. TechCrunch It says first-year collision reductions have exceeded 80%; the founder reported about 50% hardware margins, with software revenue on top. TechCrunch
• Satellite is building a Kubernetes-like software layer for running applications and AI on spacecraft and moving results across intermittent links, while leaving flight-command authority with satellite operators. TechCrunch It charges operators $4,200 per satellite per month and software vendors $10,000 plus usage fees; the founder reported over $500,000 in commercial contracts, an $8 million seed round, and an in-orbit demonstration that reduced telemetry and log data by 97% while running AI diagnostics. TechCrunch The company supports both existing satellites and pre-launch software loads, targets 50 satellite nodes this year, and is pursuing interoperability standards; its founder acknowledged that standardization could help competitors as well as Satellite. TechCrunch

• Model-market economics: Frontier labs are positioned to reach high-value markets first through scale and integrated inference; open-model businesses may take longer to build because model development, inference, and post-training are fragmented, but can serve companies that need private-data fine-tuning and ownership of model weights. The speaker says it is too early to know how the balance will settle. a16z He regards Chinese labs’ talent as comparable, with compute their main constraint; he says public evidence for distillation as a major catch-up driver is weak, estimating only a few months of benefit while stressing the large uncertainty. a16z
• Model-startup entry: The speaker argues that a good base model need not require $1 billion to build, but a new entrant still needs a model good enough for people to use and a credible release that supports its next fundraising. a16z He sees controlled pretraining as more approachable than post-training, which he describes as harder to get off the ground and still a “wild west.” a16z
• RL data and environments: The speaker sees opportunity in the environment market, but warns that buying data in bulk and filtering for what works leaves quality problems; poorly designed tasks can induce reward hacking, and vendors may prioritize world-building and revenue over model behavior. He expects quality and lab-vendor practices to take years to improve. a16z
• Open research and AI risk: Trillium’s nonprofit/open lab plans to release high-leverage RL environments and share its training stack, configurations, tasks, and trajectories so others can study model capabilities and automation; the speaker says producing environments well can require thousands or tens of thousands of dollars. a16z He says inadequate public and cross-sector cybersecurity preparation worries him more than fast takeoff, while emphasizing that AI risks are already diffusing. a16z

• AI is shortening the build cycle, making customer validation more—not less—important: Steve Blank said all eight teams in a recent Stanford class arrived with products that looked finished, rather than the early wireframes or slide-based MVPs students typically brought before; instructors found they needed to redirect teams toward understanding customer problems. The redesigned course emphasizes in-person discovery, uses AI to speed customer targeting and compare synthetic interviews with real conversations, and aims for a design partner willing to use and pay for the product and give deployment feedback. Stanford eCorner Rita’s early-stage research draft identifies evaluation as a new bottleneck as expertise becomes decoupled from production. Stanford eCorner
• Solo founding is rising in one Stanford program: Perry reports that Launchpad teams have shifted from historically averaging “2 point something” founders—usually duos or trios—to nearly one, meaning most are now single-founder teams. This is a program-level signal, not evidence of a market-wide rate. Stanford eCorner
• Shared AI tools raise the bar for differentiation, while AI-enabled small businesses may differ from VC-scale startups: Steve says enterprise companies can use the same tools to build similar products, so founders need to know what companies have already built and identify defensibility such as branding, capital, or unique insight. He distinguishes easier-to-launch AI-enabled small businesses from deep-tech ventures targeting scale; he gives a one-person business generating $5–10 million a year as a hypothetical that might still not count as a Silicon Valley startup. Stanford eCorner

• Science Corp says its Prima retinal prosthesis received EU market approval in July and has begun commercial sales in Europe. CEO Max Hodak said the company acquired a small European stimulator business at a discount and invested a further $75 million to advance the product to market. TechCrunch
• The current 2×2 mm implant provides a small central patch of sharp, high-contrast black-and-white vision, letting patients see a few letters at a time. Hodak said a next generation is being developed for human use within a year, with a goal of getting close to native acuity within five years. TechCrunch Science says the device may address different causes of photoreceptor loss if bipolar cells remain; it is starting a Stargardt disease trial and expects five more indications beyond AMD to follow. TechCrunch
• Science says the first clinical indication for its separate biohybrid neural-interface program will be an undisclosed neurodegenerative disease, and that it is also exploring stroke. Hodak described connecting brain regions with new neural channels and engineering brains as longer-term ambitions. TechCrunch
• Science uses ESM-class amino-acid language models for protein engineering; Hodak said more light-sensitive proteins could enable dimmer LEDs and ease power and heat constraints as the technology scales. The company bought a foundry in 2022, citing slow contract research and fabrication timelines as a bottleneck to faster iteration. TechCrunch
• Hodak's business thesis is that a revenue-generating company is needed to justify hundreds of millions in frontier research investment; he said restoring vision alone could make Science a generational company. TechCrunch

• Mind Robotics focuses on industrial automation, with robots intended to manipulate and assemble products ranging from cars and shoes to consumer electronics and furniture; Scaring said industrial labor represents roughly one-third of global GDP and pointed to high-cost labor markets and U.S. manufacturing reshoring as opportunity. TechCrunch TechCrunch’s interviewer described Mind as launched in November of the prior year and cited more than $1B raised; Scaring said the team had passed 100 people a few months earlier, announced Doug Field as a partner in building it, and described his time split across ventures as variable, sometimes around 50/50. TechCrunch
• Rivian has launched R2, which the interviewer characterized as existential to the company’s scaling and Scaring agreed. TechCrunch Rivian says its R2 launch processes and supply-chain management are more mature than for R1, but memory availability and pricing remain a major challenge: the company is reducing memory needs and building multiple supply paths, while higher memory prices add many hundreds of dollars to a capable vehicle. TechCrunch
• Rivian says its in-house inference platform required many hundreds of millions of dollars to develop and reduced in-vehicle inference costs by many hundreds of dollars. TechCrunch The R2 assistant uses a small in-house model to route queries between on-vehicle and cloud models; sharing user context across platforms and choosing an agent are future work, while agents will not control safety-critical functions such as braking or steering and are limited to cabin actions such as radio and HVAC adjustments. TechCrunch
• Rivian plans to launch point-to-point driving later that year and is working toward Level 4 capability over the following two years; Scaring says both point-to-point and Level 4 use the same large driving model, informed by billions of miles from Rivian’s fleet, with Level 4 adding compute and sensors. TechCrunch
• Rivian describes its Uber robotaxi effort as a $1.25B deal with a 2028 launch; Scaring estimates that 95% of U.S. miles are currently driven in personally owned vehicles and expects ownership to remain dominant even with Level 4 autonomy. TechCrunch Rivian’s $5.8B Volkswagen joint venture covers electrical architecture and software, not AI or autonomy, though Scaring sees room to expand the relationship and sell Rivian technology to other automakers. TechCrunch

Agentic AI’s ability to act in the world increases risk: the speaker described recent hacking incidents as early warnings, said their impact has so far been limited, and warned risks could grow over the next few years without more security work by industry, companies, and governments. DWS News The remarks say risks remain uncertain and call for better quantification, dynamically updated risk-proportionate certification, and balanced AI safety laws that keep pace with AI capabilities without stifling innovation; these are calls for future measures, not an announcement of enacted rules. DWS News For investors, this puts security readiness and adaptable certification among the considerations for agentic-AI ventures. DWS News

• Mosquito is extending ecological-control hardware into real-time mosquito surveillance: its founder says sensors and neural-network species identification combine with location and weather data to forecast regional disease risk; the business model pairs equipment sales with recurring dashboard/data fees. TechCrunch The company reports more than 5,000 devices across five continents, nearly $1.8 million in lifetime revenue, $3.6 million in pipeline, deployments or orders involving the UK Health Security Agency and Nigeria, and a current $5 million raise. TechCrunch A coverage caveat surfaced in Q&A: the founder said current value comes from faster results than manual collection and lab analysis, while judges questioned whether the network has enough density for citywide surveillance. TechCrunch
• Hyde is developing a thin sensor that attaches beneath a conventional watch and tracks health metrics; its app combines movement, sleep, and nutrition data for personalized AI coaching. TechCrunch The founder reported $1.5 million in preorders, a 30,000-person waitlist and 7,000 deposits, and is raising a $2.5 million seed round; the $299 tracker has an optional $6 monthly subscription for AI and nutrition features, while basic tracking is free. TechCrunch Expansion into female-health tracking remains a goal rather than a current capability: the team said it is hardware-constrained and cited 94% accuracy as a target for cycle tracking. TechCrunch
• TimeTeller uses saliva RNA with mathematical modeling and AI to reconstruct a person’s biological clock and recommend therapy timing; the company says its CE-marked product is commercially available. TechCrunch It reports reimbursement for oncology patients in Germany and cited a 10-hospital ovarian-cancer study in which timed treatment led to fivefold better tolerability, with a possible doubling of survival; its model also includes drug-specific pharma licensing, and it is raising $10 million for commercialization. TechCrunch The company says patients can purchase through its website without clinician prescription and the insurer pays directly. TechCrunch
• Intangible offers a 15-minute verbal cognitive assessment intended to predict performance under pressure and generate player scores and insights. TechCrunch The company cites a NASCAR team’s move from 18th to third after using its assessments to select pit crew, says it is testing 20,000 girls annually through a $29-per-player channel alongside annual team/org subscriptions, and is raising $1 million with $235,000 secured. TechCrunch Expansion beyond sports is early: United Airlines was described as a potential pilot discussion, and the founder cited adverse-impact testing and sports data when asked about AI hiring-law risks. TechCrunch
• Interero Bios Systems makes human intestinal organoids with multiple gut cell types, positioning them as models for testing compounds on human-relevant tissue instead of relying on animals. TechCrunch The 18-month-old company reports nearly $500,000 revenue from eight customers, tenfold manufacturing-capacity growth, more than $4 million in non-dilutive funding, and a $6 million seed round with a term sheet received that week. TechCrunch It says 80% of sales are product, while service work can generate data it hopes to license to AI companies; importantly, it reported that no customer had yet brought a drug to market based solely on its product. TechCrunch

• Andrew Ng rejects a near-term AI job apocalypse, arguing that AI automates only parts of jobs and that demand is emerging for AI-native software engineers, marketers, salespeople, and operations staff; he describes developers broadening into full-stack work as AI handles more front-end engineering. The Economist
• For early-career software hiring, Ng frames the main challenge as education lagging demand rather than a lack of jobs: he says graduates should have experience using frontier-model APIs and AI coding agents, and that AI-native interns on his team did substantial work. He also says newer data related to the entry-level prospects research cited by the interviewer was more optimistic. The Economist
• Ng expects AI progress to remain incremental rather than hit a hard limit, and says AI performing every intellectual task humans can may take decades or longer and is very unlikely by 2030. He identifies GPU servers, power plants, data-center construction, and time for experiments as bottlenecks. The Economist

The presenter described an antibody-design workflow that combines sequence and structure generation with multimer folding models, using confidence estimates to filter designs. Greylock The reported hit rate rose from a prior state of the art of about 0.1% to 18.1%; tests using 24 designs per target found hits against half of the targets. Greylock Follow-up characterization reportedly found predicted structures matched experimental results to sub-angstrom accuracy, and most designed antibodies had favorable biophysical properties. Greylock The presenter cautioned that computational checks do not replace lab validation, and said the slides were about 18 months old, with further progress since the results. Greylock

• Data-center energy use has doubled globally in five years; panelists said the grid was unprepared, and one contrasted past 18–24-month utility connections with current waits of 3–10 years, pushing some hyperscalers toward on-site generation. TechCrunch Ambrosia said its county-by-county review found median permitting delays of 6–7 years in California, about a year in Arizona, and two days in Texas. TechCrunch
• Ambrosia Energy pairs solar and batteries for off-grid power and is pursuing portable, modular compute close to generation, including for distributed inference. TechCrunch The company says its solar installation system cuts labor from about 120 person-minutes to under 8 per panel, and its in-house battery packs cost about 1.3× bare cells versus 5–7× for typical stationary-storage systems. TechCrunch Its planned 150,000-square-foot Austin factory is projected to produce 10 GWh of storage annually within two years, enough for about 300 MW of power plants per year. TechCrunch
• Bloom Energy offers an alternative on-site power approach using solid oxide fuel cells, which its representative described as quiet, non-combustion and water-free; he said Oracle dropped gas turbines after community opposition and planned to use 2.5 GW of Bloom fuel cells at its campus. TechCrunch
• A panelist described energy hardware as having a deep “valley of death”: projects can cost billions and are generally financed by banks rather than VCs, while risk-averse lenders can constrain innovation; he also said solar robotics had struggled. TechCrunch

• Eric Ries argues that faster building does not remove customer learning as the startup bottleneck; he warns that AI-generated customer stand-ins reflect their training distribution and may steer novel products toward familiar patterns. Stanford eCorner
• For evaluating early-stage progress, Ries says to prioritize customers, revenue, and evidence of solving a real problem over code volume or polish. He defines an MVP as the minimum functional product that lets a customer complete a task, and says AI has not fundamentally changed that principle. Stanford eCorner
• On mission-driven startups, Ries says companies trusted by customers, employees, and communities have a competitive advantage; he advises founders to present mission as part of how they intend to maximize the company’s outcome, rather than ask investors for permission, and argues investors will support it when they believe it can make money. Stanford eCorner

• Helion: The company said it had raised more than $1.5 billion at a $15 billion valuation. TechCrunch It targets 2028 for its first Orion plant and the end of the decade for grid electricity; it said buildings were built and much of the permitting obtained. TechCrunch Microsoft signed a 50-MW power purchase agreement with Helion in 2023. TechCrunch
• Commonwealth Fusion Systems (CFS): CFS said its Spark test device was on track to produce more power than it consumes “next year”; the interviewer described that milestone as two years later than the 2020 forecast. Its first commercial ARC plant targets grid connection in the early 2030s and has a 200-MW Google PPA. TechCrunch
• Execution and policy: CFS cited sourcing and coordinating fusion-specific components as a challenge. TechCrunch Helion said reliance on global pulse-capacitor supply led it to plan a U.S. manufacturing plant and a 60x production scale-up. TechCrunch Helion said the U.S. fusion regulatory issue was resolved and its first facility licensed; CFS said several other countries had adopted similar fusion-specific policies. TechCrunch
• AI applications: CFS said it was the only private company selected to lead a government Genesis Award effort developing an AI digital-twin “co-pilot” to guide tokamak operators, rather than pilot the machine itself. TechCrunch Helion described AI uses in software and manufacturing, including a reported 30x speed-up in writing firmware and code for automated systems. TechCrunch

• Cerebras CEO Andrew Feldman said AI-compute demand and data-center construction are growing together, while arguing that efficiency improvements are complementary and needed across power delivery, networking, compute hardware and models. He also acknowledged that the industry had not engaged local communities effectively, particularly in the U.S. TechCrunch
• The interviewer identified OpenAI and AWS among Cerebras’s major customers, alongside neoclouds and companies including Cognition, Lovable, Block, Figma and GSK. TechCrunch Feldman said Cerebras lacked data-center capacity for the wider next-generation deployment OpenAI wanted; he described the arrangement as Cerebras building data centers, installing its compute and selling OpenAI tokens, and said the multi-year relationship was going fine. TechCrunch
• Feldman said Cerebras spent about $8 million a month for 18 months while it was still unable to solve key technical problems. He said an early chip lasted eight seconds against a 10-year durability goal, and described repeated testing and failure analysis before the technology reached production. TechCrunch

• The hosts say lab-grown diamonds account for 60%+ of the market (without specifying the measure), cost 80–90% less than natural diamonds of the same quality, and are visually and chemically indistinguishable to consumers, though a jeweler may detect them with testing. They describe De Beers first disparaging lab-grown stones, then selling its own low-priced lab product to reinforce a distinction from natural diamonds. My First Million
• For data-center infrastructure, the hosts suggest studying De Beers’ perception-shaping playbook, saying the industry seems unpopular with roughly 70% of people; that figure is their estimate, not cited polling. My First Million

• GPU-based data-center racks’ peak power rose from 10 kW to 200 kW over five years, and panelists said designs for 4 MW racks are emerging—power levels they described as not cost-effective to serve with copper. TechCrunch
• A startup developing superconducting transmission lines says its conductors can carry 10× more power in the same right-of-way. After four years focused on that product, it pivoted three years ago to lower-voltage, behind-the-meter and data-center deployments as a beachhead toward utility applications, citing utilities’ risk aversion and long timelines. TechCrunch
• Weave Grid’s approach coordinates EV charging and other loads to reduce simultaneous peaks; its representative said the company is live in 25 or 26 states and has integrations that can reach roughly 10 million cars and batteries. The representative also said utilities are asking how the company can scale and bring gigawatts online faster than new power plants or substations. TechCrunch
• AI data-center demand is driving grid investment but also straining local distribution and competing with household electrification. Panelists said household flexibility can sometimes be deployed sooner than new power plants or transmission, and suggested data centers could fund household solar, batteries, EVs or heat-pump water heaters that aggregate into virtual-power-plant capacity. TechCrunch

Greylock's recruiting advice for founders hiring beyond their own networks is to go beyond explaining what the company is building: explain why the founders chose this market and help candidates see how working with the team could benefit their longer-term careers. Greylock For ambitious companies, a founder's story should connect the decision to build with a genuine catalyst and why they are suited to pursue it; it need not rest on a personal connection, but authenticity can help attract strong talent. Greylock

• Inertia Enterprises is led by Jeff Lawson, Twilio co-founder and CEO through 2024, and includes LLNL ignition-experiment designer Annie Crater and a co-founder who ran Stanford’s linear accelerator and designed a fusion power plant. The company says it licensed 200 LLNL patents, established three joint research agreements with the lab, and formed the largest public-private partnership in the national-lab system’s history. TechCrunch
• Inertia is pursuing laser-driven inertial confinement, building on LLNL’s December 5, 2022 ignition result. Its commercialization thesis is to replace the experiment’s enormous vacuum-tube laser with a smaller, more powerful and efficient diode-based system, while automating fusion-target production. TechCrunch
• The scale-up challenge remains substantial: Lawson described the existing laser as about 0.5% efficient, with targets costing about $500,000 each and produced at six per year; a power plant would need roughly one million targets daily. TechCrunch
• Inertia reported $450 million in Series A funding commitments, a 50,000-square-foot facility near LLNL, and about 75 employees, up from five at the beginning of the year, working on laser and target development. It also said it had recruited manufacturing expertise from Apple and Waymo. TechCrunch
• The company’s stated roadmap is to break ground on a fusion pilot plant in 2029, achieve first light around 2035, and have three plants operating or under construction by the end of the 2030s. Lawson identified demonstrating net gain at a private facility and putting its first electron on the grid as a key validation milestone, while saying fusion is a 2030s-and-beyond effort, not a near-term answer to rising power demand. TechCrunch

Billy Evans, whose background includes photonics work at Luminar on lidar for autonomous vehicles, says his company Hemanthes is building a new layer of biological sensors; he frames the opportunity around diagnostics being too expensive and inaccessible, with few ambitious efforts in the space after Theranos failed. All-In Podcast The company uses Raman spectroscopy, a technique already used in research labs, and aims to make the equipment—described as costing millions of dollars and being bulky and complicated to use—cheaper and more widely distributed. All-In Podcast Its stated goal is to generate rich biological datasets for AI to find correlations across disparate signals. All-In Podcast

• Agentic payments face trust and recourse gaps: panelists said institutions lack a clear refund or dispute process for agent purchases, and one payment-system executive said merchants may block agents because they bear chargeback risk. TechCrunch Agent identity and history, AML checks, and liability when agents make mistakes also remain unresolved. TechCrunch A panelist’s shopping-agent trial failed to beat Amazon’s three-click cat-food order, though he saw more potential in higher-consideration purchases such as flights, hotels, and complex appliances. TechCrunch
• The startup opportunity spans both new and incumbent payment infrastructure: an Emergence Capital seed/Series A investor said she is tracking stablecoin-native financial institutions as well as companies improving card rails with fraud and dispute checks. TechCrunch A Circle representative said agent-to-agent commerce is still small but claimed 98% of it uses stablecoins, and described programmable stablecoin treasury and vendor-payment flows as already operating in B2B. TechCrunch The counterview is that traditional rails serve consumer payments well and RTP/FedNow may be better suited to point-of-sale payments, while stablecoins allow fintechs to settle globally without directly accessing those bank-only systems. TechCrunch
• In discussing the GENIUS Act, panelists said its restriction on interest for stablecoin deposits shifts the consumer proposition toward faster settlement; the counterargument was that users can move stablecoins into and out of yield-bearing money-market products around the clock. TechCrunch