No Priors

Sarah Guo × Dane Knecht

Vulnerabilities once seen annually now appear weekly; Knecht expects security to consume most engineering time for 18 months.Cloudflare counters with simpler code, independent reviews and free major-vulnerability protection; its red team may grow from one to four or five by next year’s end.Agent wallets, a stablecoin and its gateway test CPM-priced micropayments, though readiness remains uncertain.

Sarah Guo × Elad Gil × George Kurtz

AI is flattening attackers’ hierarchy: Kurtz cites 29 minutes on average to move beyond initial compromise, with a fastest observed 27 seconds.CrowdStrike’s AIDR traces agent identities, subagents, tool calls and data access; Kurtz sees agent security as potentially larger than endpoint protection.Kurtz says a leading cybersecurity frontier model probably costs about $10 versus CrowdStrike’s about 3 cents; “low and slow” attacks may be harder to identify.

Sarah Guo × Elad Gil × Greg Brockman

Brockman says AI is now creating reusable mathematical insight, with almost 400 important open problems solved through unexpected combinations of ideas.Drug development offers the clearest disruption case: a roughly 15-year, $1.5 billion process could accelerate in literature review, simulations, experiments, and regulatory writing, but biological data and long-term outcomes remain constraints.As code approaches near-zero cost, Brockman predicts cybersecurity spending should “far exceed software spending,” while OpenAI’s defender’s window and recurring model-led security testing remain catalysts to monitor.

Sarah Guo × Elad Gil × Misha Laskin

Reflection AI’s Beam makes a differentiated open-model wager: frontier intelligence requires a full-stack lab, evidenced by growth from roughly 30 to 300 people and a 500-billion-parameter model with 23B active parameters.Its claimed three-to-fourfold reasoning efficiency, alongside catch-up costs potentially reaching around $10 billion next year, could support enterprise shifts from renting closed models to owning open deployments once annual spend exceeds roughly $100 million.

Walter Goodwin × Sarah Guo

Fractile is betting frontier inference will be constrained by memory bandwidth and cost, not compute, shifting from SRAM to a DRAM architecture expected to be fully operational in the second half of next year.Goodwin claims 25 times more bandwidth per chip than an HBM-based design, potentially making sparse MoE models economical, while its 150-person team targets a three-to-six-month lead; foundry cycles, ramps, and three-to-five-year amortization remain constraints.

Sarah Guo × Michael Lee

Sequence Holdings is pairing frontier engineering with incumbent advantages, partnering with the Dell family office on its $7.7B acquisition of Baldwin.At BankSouth, average consumer-loan underwriting volume fell 94%, timelines dropped from 30 days to 11, and Q2 loan volume doubled versus Q1 without changed standards.Whether Atlas can repeat that operating leverage depends less on code than on the more difficult human engineering problem.

Sarah Guo × Stefano Ermon

Diffusion inference is becoming a commercial signal: at GPT-2 scale, it matched autoregressive perplexity using the same data and parameters while generating text roughly 10x faster.Mercury extends the wedge to production: Ermon says OpenCall moved from Cerebras to NVIDIA GPUs for comparable speed, lower cost, and higher quality, while Inception’s proprietary stack and frontier-intelligence gap remain key tests.

Brian Armstrong × Elad Gil

Coinbase says roughly 76% of agentic transactions are below 30 cents, making card rails uneconomic and supporting self-custodial wallets plus the X42 protocol.With 88% of revenue now non-Bitcoin trading, prediction markets reaching a $100M revenue run rate and tokenized stocks could broaden the exchange, while the SEC path for U.S. availability remains unresolved.

Rene Haas × Elad Gil × Sarah Guo

Arm moved from IP licensing into physical silicon with the Arm AGI CPU after Meta sought a general-purpose agentic CPU no one else could provide.AI reaches 80% to 90% of engineers, while Arm’s documentation and test benches may make proprietary IP more trainable; supply may remain constrained for 3 to 5 years, with data-center construction a possible next bottleneck.

Elad Gil × Ofir Ehrlich × Gonen Stein

Models and compute have almost zero switching cost; accumulated enterprise data is becoming the moat, underscored by Google’s $10M purchase of bankrupt Spirit Airlines’ data.Eon maps and classifies siloed data, controls sensitive access, and exposes it to AI without compromising production or compliance, while agentic permissions create extreme-velocity ransomware risk and most companies still do not use AI.