We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
1. Funding & Deals
Apollo Atomics raised a $31M seed for high-power, truck-deployable reactors. The company says the round comprises $26M of equity led by FCVC, with Y Combinator, TeleSoft Partners, Alumni Ventures, Pelion Venture Partners, Duke Capital Partners, Robinhood Ventures, and Nucleation Capital participating, plus $5M of equipment financing from Oak Ridge Financial. It claims its design can deliver more than 80 MW on a truck without exotic or scarce fuel, unqualified materials, or high regulatory uncertainty.
Apollo is running a low-power demonstrator at MIT, building a full-scale 1 MW prototype under commercial conditions, and working with the U.S. Nuclear Regulatory Commission toward commercial deployment in 2028. It is also building software for autonomous reactor control, licensing support, and reactor modeling. The investment case therefore spans hard-tech execution and software-enabled deployment; the 2028 target is the key schedule to diligence.
AnySignal won a U.S. Space Force contract to demonstrate satellite-to-satellite communications on orbit. The system is intended to move sensor data between satellites in near-real time without routing through the ground, extending the company’s work on resilient MILSATCOM and decentralized mesh networking for contested environments. This is a useful non-dilutive customer-validation signal for resilient space infrastructure.
2. Emerging Teams
a16z is turning “Borderless Founders” into an explicit early-stage sourcing thesis. Its argument is that founders with one foot in their home country and one in Silicon Valley can land early customers through local and diaspora relationships, while accessing home-country talent before it becomes obvious to the broader market. Pit, a seed-stage Stockholm AI platform, landed H&M and Stena Metall as first customers; CEO Adam Jaffer had previously founded Voi Mobility. Supersonik founder Daniel Carmona used the Spanish diaspora to secure Salesforce as the company’s first major contract.
The playbook is operational, not merely cultural: a16z maps country nodes, hosts founder dinners, maintains in-person presence, builds diaspora communities, and says it invested at seed in two founders who had previously built Kry and Voi. It reports that 44% of a16z Apps’ early portfolio companies feature international founders, split evenly between U.S. and foreign headquarters. For seed investors, geography and diaspora reach can be underwriting inputs for customer acquisition and recruiting—not just biographical color.
Chai Discovery added Bristol Myers Squibb to its reported biopharma customer roster. Chai says BMS joins Eli Lilly, Novartis, Pfizer, and argenx in using its models for therapeutic discovery since Chai-3 was rolled out to partners at the start of the year. That is a strategic-validation signal for scientific AI, though the announcement provides no financing or outcome data.
3. AI & Tech Breakthroughs
Harvey’s Tenet is the clearest current example of specialized post-training producing both quality and cost gains. Harvey says Tenet uses Kimi K3 as its base, was post-trained with Fireworks for long-horizon legal work, and learned from synthetic data, public legal data, and human expert data. On LAB holdouts, it completed almost twice as many tasks as base Kimi K3 and 20% more LAB Contracts tasks, while increasing all-pass rates by 9 and 2 percentage points; it reports state-of-the-art performance on LAB Contracts and second place on LAB. Reward shaping targeted efficient tool use and reasoning, allowing Harvey to improve performance while keeping cost stable.
The broader result is a stack rather than a fine-tune. On an M&A diligence task involving up to 80M tokens of documents, Harvey’s recursive-language-model harness lifted GLM-5.2 from a 46.1% criteria-pass rate to 60.1% after post-training. In Review Table, the post-trained model improved answer quality by 3.6 points and citation quality by 12.1 points at roughly one-tenth the cost per cell. In a separate Engram partnership, parametric memory improved criteria pass rate by more than 15%, cut tokens by 58%, reduced cost per query by 90%, and raised Harvey’s intelligence-per-token score to 190.8 versus 129.3 for its best frontier configuration. The investment implication is that domain environments, harnesses, memory, and evaluation may be more defensible than access to a general model.
Model access is becoming an orchestration surface. Perplexity’s Agent API exposes 41 frontier models across nine providers through one endpoint, with web, finance, fetch, and sandboxed-code tools. RouteLLM advertises caching and access to more than 150 models, routing simple turns to open-source models and complex, long-running tasks to frontier models. The signal is a continued shift toward task-level model selection and tool execution as developer primitives.
Memory is emerging as a standalone agent-infrastructure category. Chroma’s Foundation research preview says it builds self-improving memory from agent sessions. The research-preview qualification matters: this is an infrastructure direction to track, not yet evidence of production traction.
4. Market Signals
Owner’s operating data shows how agentic vertical SaaS can invert conventional product metrics. Owner reports that more than 83% of new customers now start in its AI product, up from 0% two years earlier, with faster growth in 2025 and year-to-date 2026 than in 2024 and the company approaching $100M ARR. Its stated advantage is an opinionated product that enforces one system across customers and accumulates outcome data about what drives restaurant sales. CEO Adam Guild argues that dashboard logins can become a failure signal: if customers must return to fix what the software did, the agent failed. For early-stage underwriting, ask whether an agent reduces manual intervention while generating proprietary outcome data, even if DAU or MAU declines.
AI labs are pulling senior engineering talent out of incumbents despite unusually large retention offers. A report based on interviews with Meta engineers says the company offered IC6/IC7 staff and principal engineers $400K to $1M+ in retainer equity, vesting over three years; $1M+ offers were associated with Anthropic or OpenAI offers. Of three engineers who received $1M+ counteroffers, two joined Anthropic and the third left a month later after initially accepting Meta’s grant. The same account describes a long-tenured AI infrastructure engineer choosing a startup after its founder materially increased the equity package and made the role feel more consequential. This is interview-based evidence rather than a market-wide survey, but it raises the value of founder-led recruiting and meaningful technical ownership for seed-stage teams.
Compute scarcity is becoming a company-risk variable. Conviction founder Sarah Guo warned that many strong AI companies could become desperate because growth is limited by compute. The warning is not quantified, but it is a direct investor signal to diligence capacity access, financing runway, and whether a company’s product can grow under constrained inference supply.
Deployment policy is moving quickly in autonomous mobility. TechCrunch said Nevada permits would allow up to 8,000 robotaxis to be deployed over the next 12 months; Jason Calacanis characterized the change as a rapid reversal from Nevada’s earlier resistance to ride sharing. This is a regulatory tailwind signal, not evidence that fleet economics or operational reliability are solved.
5. Worth Your Time
- Read — Harvey’s “Update on our post-training effort”. The primary technical account of Tenet, recursive-language-model harnesses, reward shaping, and firm-specific memory.
- Watch — YC Paper Club: data, benchmarks, and multilingual pre-training. YC frames the session around old training and evaluation assumptions breaking as models scale, with talks on data, agent benchmarking, production diffusion language models, and multilingual scaling laws.
- Read — The Pragmatic Engineer’s “The Pulse: We need to talk about migrations with AI”. Asana completed a large Enzyme test migration in two weeks with AI, while Airbnb and Uber report similar experiences; the same issue questions whether legacy-vendor rankings retain influence as AI-native tools emerge.
- Read — a16z’s “Rise of the Borderless Founder”. A practical sourcing and portfolio-support playbook for country networks, diaspora distribution, and repeat founders.
a16z Borderless Founders theme — Over 40% of a16z Apps team investments in the last two years went to international founders, half headquartered outside the US; the bet is led by Angela Strange and Gabriel Vasquez . The thesis: AI opens every market but concentrates the epicenter in the Bay Area, and country diasporas beat elite alumni networks — borderless founders keep one foot in their home country and one in Silicon Valley, gaining early customers from large home-market companies and a diaspora that answers calls before traction warrants it .
Portfolio proof points: ElevenLabs (Poland) has become that country's national AI champion — the Polish government joined its cap table and opened an AI lab in Poland to spur more AI companies from its 45,000 annual technical graduates . Supersonik (Spain) landed Salesforce as its first design partner via the Spanish diaspora — founder Daniel Carmona connected with Salesforce's CRO, who studied at the same Barcelona university . Addi (Colombia) was Strange's first check and now serves a quarter of the country, recruiting Capital One credit talent to Bogotá .
Borderless founder advantages: customers, brand, and talent . Pit (Stockholm, seed-stage AI platform) closed H&M and Stena Metall as first customers; CEO Adam Jaffer previously founded Voi Mobility . Segura (Brazilian insurance AI) recruits from lesser-known university robotics teams . More than half of America's billion-dollar companies were founded or co-founded by immigrants, and 44% of a16z Apps early portfolio companies have international founders, split evenly between US and foreign HQ .
Sourcing playbook: a16z backs repeat borderless founders — it invested at seed in two founders who previously built Kry and Voi, and pre-seed in Fernando Gadotti, who went on to co-found Tako with Rappi co-founder Sebastian Mejia to reduce bureaucracy in Brazil starting with payroll . Its Borderless Founder dinner series has hosted 600+ founders over five years, pairing established founders — Sebas Mejia (Rappi), Guillermo Rauch (Vercel), Pedro Franceschi (Brex), Fredrik Hjelm (Voi), Kristo Käärmann (Wise) — with the next generation; a Ben Horowitz dinner in Stockholm led government representatives to double down with ministerial access and trade diplomacy . Full playbook discussion in an a16z podcast with Strange and Vasquez: https://www.youtube.com/watch?v=0t3TpJXa5-A.
a16z is building an investment thesis around 'Borderless Founders' — international founders with one foot in their home country and one in Silicon Valley — who gain compounding advantages in talent, customers, and brand that accelerate preferential attachment .
- Concrete early-stage examples: Pit (seed-stage AI platform in Stockholm; landed H&M and Stena Metall as first customers; CEO Adam Jaffer previously founded Voi Mobility) ; Supersonik (founder Daniel Carmona used the Spanish diaspora to secure Salesforce as first major contract) ; ElevenLabs (Polish government joined the cap table and opened an AI lab in Poland) ; Segura (Brazilian insurance AI hiring from lesser-known university robotics teams) ; Tako (Brazilian, co-founded by Fernando Gadotti and Rappi co-founder Sebastian Mejia, reducing bureaucracy starting with payroll; a16z backed Gadotti's previous company at pre-seed) ; Slash (CEO Victor Cardenas, now unicorn, attended a Borderless Founder dinner 3 years ago) ; Factor (German-led, introduced to a large LatAm bank as first pilot customer) .
- a16z's playbook: map country nodes and top founders, host dinners with luminaries like Sebas Mejia, Guillermo Rauch, Pedro Franceschi, Fredrik Hjelm, and Kristo Käärmann, maintain in-person presence, and invest in repeat founders (two dinner attendees who built Kry and Voi were invested in at seed for new companies) ; build diaspora communities ; provide 'go-direct' media support ; broker cross-border customer introductions like Factor to a LatAm bank .
- Key data: 44% of a16z Apps early investments feature international founders, split evenly between US and foreign HQ; more than half of US billion-dollar companies were founded/co-founded by immigrants .
- Angela Strange (a16z GP): 'Borderless Founders have more levers at their disposal'; in the 'fiercest talent market' of her career, they can tap overlooked talent (an AI insurance company in Brazil recruiting from a little-known university), which starts a 'talent flywheel'; on customers, 'nobody wants to be the first bank or insurance company,' but the diaspora network lets them land a large bank/insurer abroad quickly .
Apollo Atomics raised a $31M seed round: $26M in equity led by FCVC, with participation from Y Combinator, TeleSoft Partners, Alumni Ventures, Pelion Venture Partners, Duke Capital Partners, Robinhood Ventures, and Nucleation Capital, plus $5M in equipment financing from Oak Ridge Financial . The company says it is the only nuclear firm delivering >80 MW reactors on a truck, without relying on exotic/scarce fuel, unqualified materials, or high regulatory uncertainty . With Drew Walker, it has assembled a talent-dense team, is running a low-power reactor demonstrator within MIT, building a full-scale 1 MW reactor prototype under commercial conditions, and working with the U.S. NRC toward commercial deployment in 2028 . It is also building software for autonomous reactor control, licensing support, and reactor modeling . Individual investors include Paul Graham, Evan Meagher, Philip Johnston, Ray Rothrock, and Matteo Franceschetti . Garry Tan framed the round by saying, "YC is the YC for hard tech" .
YC President & CEO Garry Tan posted "YC is the YC for AI Researchers" while sharing YC's Paper Club event on data, signaling YC's push to position itself as the center of the AI research community .
The Paper Club session (video) focused on how "old assumptions about training and evaluation are starting to break" as models scale . Presentations included: @FrancoisChauba1 on why a whole night about data, @vincentsunnchen on "The Art & Science of Benchmarking Agents", @volokuleshov on "Inception: Diffusion Language Models for Production", and @ShayneRedford on "ATLAS: Practical Scaling Laws for Multilingual Models" .
Artie, a real-time data replication startup, launched a free plan so anyone can run real-time replication without building and operating the infrastructure themselves . The company frames the hard part of moving data as keeping it correct and current, and recovering when pipelines restart, schemas change, or data falls behind ; signup is at http://go.artie.com/free-plan. Dalton Caldwell (@daltonc), founder and partner at Standard Capital and Partner Emeritus at Y Combinator, publicly shared and endorsed the launch .
A post by @amasad (speaking from Replit's early history) announced a partnership with OpenAI, calling it "long overdue" and linking to an official Replit announcement for details . The post also recalls that before Replit was in YC, Paul Graham asked Sam Altman to recruit the team, signaling early backing from YC leadership and Sam Altman . Vinod Khosla publicly endorsed the partnership with a short celebratory post: "Yes, itrs time!" .
Bill Ackman and Neri Oxman are founding the Ackman Oxman Institute (AOI), a non-profit brain research, rehabilitation, recovery, and longevity institute, anchored by a ~$400M gift of 10M Pershing Square shares . The AOI will run its own venture funding to seed, assist, and spin out companies, with focus areas including neurosurgery, neuroscience, rehabilitation, BCI and device development, human trials, and life extension programs . Ackman highlights BCI work underway at Neuralink, Precision Neuroscience, Science, Synchron, Nudge, and others, and says AOI will assist existing companies in the space; AOI also aims to be at the forefront of the brain–AI interplay . The institute's 680k sq ft lab campus will be larger than Rockefeller University's; board includes Regeneron CEO George Yancopoulos and Nobel laureate James Rothman, and a CEO is expected by October .
An anonymous X user (@wolframs91) put forward an unverified theory that Opus 4.6 was the last Anthropic Opus generation heavily used by Anthropic employees, and that Opus 4.7, 4.8, and 5 were trained via RLAIF using an internal 'Mythos' model's teachers . The user suggests this could explain why 4.6 does not 'report back like a robot' and why Opus 5's coding style is criticized , while explicitly disclaiming certainty ('I have no idea if this is true') . Garry Tan quoted it as an 'Interesting violation of "you should dogfood your own product"' . No evidence is provided; treat as speculation about Anthropic's internal model use and training approach.
Y Combinator's Paper Club session focused on data-centric AI frontiers, framed around old assumptions about training and evaluation breaking as models scale across tasks and languages. Talks covered: why data matters (@FrancoisChauba1), the art and science of benchmarking agents (@vincentsunnchen), diffusion language models for production (@volokuleshov, "Inception"), and practical scaling laws for multilingual models (@ShayneRedford, "ATLAS") . The full session video is linked in YC's follow-up post .
Paul Graham argues that the winning approach to capitalism is focusing on making customers' lives better rather than the 'socialist caricature' of squeezing every penny; the latter yields at most 2x returns, while creating new things can yield 10x . He clarifies that those implementing the caricature are mostly opportunists and bean counters, not the big stars, who make money by creating wealth .
Andrew Chen (@andrewchen) posted the phrase "agent led growth" with no elaboration .
Paul Graham frames startup resilience as the "cockroach" concept: he says YC's mascot would be a cockroach, a trait he values in founders . He shared a 2008 essay (written during the financial crisis) explaining the origin of the term, calling Airbnbs the quintessential cockroaches — a quality that served them well in 2009 and during covid . The essay is at https://paulgraham.com/badeconomy.html.
YC CEO Garry Tan hosted designer John Maeda at YC, stating design is more important than ever in the time of AGI and that "knowing what to build" is the primary key when you can build anything .
Perplexity announced the Agent API, offering developers access to 41 frontier models across 9 providers through one endpoint , with built-in tools including web search, finance search, fetch, and sandboxed code execution for multi-model agent workflows . CEO Aravind Srinivas frames this as a full-fledged AI developer platform: access to various frontier and workhorse models plus tools for deploying them in production workloads .
Meta is losing senior engineering talent to AI labs and startups despite newly introduced retention grants. Meta now offers $400K–$1M+ retainer equity (vesting 3 years) to resigning IC6/IC7 engineers, with $1M+ grants aimed at those holding Anthropic or OpenAI offers; IC4/IC5 engineers report no such offers . The tactic is failing: of three engineers offered $1M+ counter-offers (vesting 4 years) to stay, two still joined Anthropic and the third initially accepted but left a month later, forfeiting the grant . A long-tenured AI infra engineer chose Meta over a high-growth AI startup and Google, but reversed after the startup founder significantly increased equity, resigning to join . Anthropic and OpenAI are the main destinations for ex-Meta engineers; both run secondary equity sales and their IPOs are expected within 3–12 months . Separately, Meta released Muse Spark, an AI model now ranked 7th most capable (tied with Grok 4.5) and ahead of Google's Gemini . In AI code modernization tools, Gartner ranked AWS, Microsoft, and IBM above Anthropic, Cursor, and OpenAI, likely because the former pay Gartner while AI-native vendors refuse the 'Gartner tax' .
- Harvey's research update: post-training focus on frontier legal intelligence with open-weight models, plus tooling for law firms to own their intelligence .
- Tenet, Harvey's first post-trained model (Kimi K3 base + Fireworks), targets long-horizon legal work and shows promising performance/cost results .
- Tenet completes ~2x held-out LAB tasks and +20% LAB contracts vs base, with SOTA on LAB Contracts; gains transfer to Mercor APEX Agents (corporate law) and Crosby's Redline Bench .
- Cost-efficiency via reward shaping for efficient tool use/reasoning, cutting inference tokens; open-weight token prices cheaper .
- Training: ~1,750 agentic legal environments, GSPO with rank-64 LoRA, ~150 NVIDIA B300 GPUs over 2 months; LLM-as-a-judge grading (Kimi 2.6 as best judge) .
- M&A diligence: RLM harness (root agent + delegated sub-agents) lifts GLM-5.2 post-trained to 60.1% LAB Diligence criteria pass vs 46.1% base and 43.8% best baseline .
- Review Table: post-trained GLM-5.2 with Applied Compute gains +3.6 answer quality and +12.1 citation quality at ~1/10 cost per cell .
- Firm knowledge: Engram-partnered Qwen3.8-27B with parametric memory improves criteria pass >15%, cuts tokens 58%, cost/query -90%, triples intelligence-per-token (190.8 vs 129.3 best frontier) .
- Partners include Fireworks, Engram, Baseten, Applied Compute, NVIDIA, Mercor, Snorkel AI .
- Sarah Guo highlights the work as a blueprint for specialized intelligence: "own your intelligence" .
- A founder with a prior exit — a news/media aggregator for high-growth industries that was acquihired, with the founder and team staying 3+ years — has started a venture in public-sector data, initially a side project that is becoming his main focus; the team's other side has heavy public-sector experience .
- The product is a Spokeo/WhitePages-style service over a claimed ~2B possible end reports: B2C consumers pay to unlock full reports; B2B customers get a search dashboard or API for white-label/co-branded results, with long sales cycles . The founder previously indexed ~350k of ~90m pages on his prior aggregator (~33k backlinks) and is weighing mass indexing of millions of report pages; a commenter advises against bulk indexing as the first goal — start with a few hundred/thousand templates tied to real search intent and set a quality threshold before pages become indexable .
Nevada permits would allow deployment of up to 8,000 robotaxis over the next 12 months . Commentator @Jason frames this as Nevada going from blocking ride-sharing for a decade to speed-running 8,000 AVs in a year , signaling a fast-moving regulatory tailwind for autonomous ride-hailing and robotaxi entrants.
- AI-driven code migrations gain traction: Asana completed a migration off the Enzyme testing framework using AI in two weeks—work that would otherwise have been delayed for years; Airbnb and Uber report similar AI-assisted migration experiences, indicating AI is a strong fit for framework migrations and a growing trend in AI-assisted software engineering .
- AI startups disrupting Gartner's influence: Gartner's Magic Quadrant for AI code modernization tools ranks AWS, Microsoft, and IBM above AI-native vendors Anthropic, Cursor, and OpenAI, likely because the former pay Gartner fees while AI startups refuse, signaling a shift in how enterprise software purchasing decisions are influenced .
President Trump criticized Texas Gov. Greg Abbott's data center pause as "a mistake," saying data centers are "tremendously important for the economics" and could be "bigger than oil" for Texas . Jason Calacanis interprets the pause not as a ban but as Abbott asking developers to disclose energy needs and plans before breaking ground .