# What AI leaders said in October 2026 Markdown copy of https://realalpha.app/q/numbers/2026-10/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy. [All questions](https://realalpha.app/q/) 296 answers to the big questions about AI, from the long interviews that came out in October 2026 and that we cut. Each person counts once per question, by their newest answer. Published Nov 1, 2026 · [Every answer as CSV](https://realalpha.app/questions.csv) ## October in five lines - GPU shortage. 10 of 12 say compute is still short (Sundar Pichai, Matt Garman and Evan Conrad among them). 2 say the shortage is easing. None say a glut is coming. - Power. 8 of 9 say power is the bottleneck (Aravind Srinivas, Satya Nadella and Misha Laskin among them). 1 says it’s tight but solvable. None say power isn’t the problem. - AI bubble. 7 of 12 say it isn't a bubble (Sundar Pichai, Jensen Huang and Mark Zuckerberg among them). 5 say it partly is. None say it is a bubble. - Agents. 27 of 28 say they’re ready for some work (Jake Stauch, Sundar Pichai and Bernardo Aceituno among them). 1 says they’re not ready yet. None say they’re ready now. - Jobs. 15 of 23 say jobs will change more than vanish (Marco Palladino, Jad Naous and Han Wang among them). 7 say AI will take over jobs. 1 says some work stays with people or grows. ## [What will AI look like in the next few years?](https://realalpha.app/q/ai-predictions/) 26 of the 41 people who answered it in October look a year or two out. 8 look a decade or more out. 7 look about five years out. ### In a year or two · 26 - **Sam Altman** (CEO, OpenAI) said on Cheeky Pint on Oct 29 that teams of agents may scale well and help keep one another in check. [Hear it at 11:14](https://realalpha.app/s/agent-teams-may-improve-both-capability-and-alignment/) - **Marco Palladino** (CTO, Kong) said on MTS on Oct 20 that people will increasingly design agent workflows while agents handle operational work. [Hear it at 19:49](https://realalpha.app/s/workers-will-shift-from-doing-tasks-to-orchestrating-agents/) - **Tiago Sada** (CPO, World) said on MTS on Oct 20 that online services will increasingly be accessed through agents acting with proof delegated by specific users. [Hear it at 5:29](https://realalpha.app/s/agents-could-prove-they-act-for-a-specific-human/) - **Willie Williams** (Head of Platform, Every) said on AI & I on Oct 7 that work will use one shared agent per company, while personal agents will serve people at home. [Hear it at 18:04](https://realalpha.app/s/work-will-run-on-company-agents-while-personal-agents-stay/) - And 22 more: [Woodson Martin](https://realalpha.app/s/ai-is-making-decades-old-system-migrations-feasible-in/), [Tibo Sottiaux](https://realalpha.app/s/most-internet-actions-will-soon-be-taken-by-agents/), [Pierre Stock](https://realalpha.app/s/mistral-says-its-models-speed-senior-researchers-up-two-to/), [Ivan Burazin](https://realalpha.app/s/real-world-ai-will-become-more-visible-in-san-francisco/), [Tanishq Mathew Abraham](https://realalpha.app/s/cancer-ai-tools-could-reach-clinical-trials-within-a-year/), [Sundar Pichai](https://realalpha.app/s/ai-should-adapt-to-people-but-screens-will-still-have-a-role/), [Josh Gao](https://realalpha.app/s/constrained-jobs-will-come-before-household-robots/), [Bernardo Aceituno](https://realalpha.app/s/enterprise-agents-will-spread-across-apps-and-act/), [Dmitri Dolgov](https://realalpha.app/s/ai-will-speed-simple-builds-while-safety-critical-work-stays/), [Sigil Wen](https://realalpha.app/s/todays-frontier-models-will-run-on-devices-in-six-months/), [Jerry Wu](https://realalpha.app/s/rl-environments-roughly-double-in-complexity-every-6-8/), [Bill Ackman](https://realalpha.app/s/ai-could-help-restore-lost-brain-function/), [Gavin Baker](https://realalpha.app/s/gpu-supply-is-tight-despite-the-capacity-coming-online/), [Aghi Marietti](https://realalpha.app/s/enterprise-ai-is-spreading-across-models-as-token-use-surges/), [Marc Andreessen](https://realalpha.app/s/ai-is-widening-the-gap-between-startups-and-incumbents/), [Vlad Tenev](https://realalpha.app/s/agent-led-product-work-depends-on-automated-testing/), [Pat Gelsinger](https://realalpha.app/s/ai-speeds-chip-design-but-fab-memory-and-power-still-bind/), [Jay Madheswaran](https://realalpha.app/s/routine-legal-back-and-forth-may-move-between-agents/), [Scott Kirby](https://realalpha.app/s/ai-could-cut-denvers-summer-weather-delays-by-up-to-75/), [Dave Blundin](https://realalpha.app/s/ai-is-cutting-the-cost-of-replacing-legacy-software/), [Jake Cooper](https://realalpha.app/s/agent-growth-could-make-compute-the-next-infrastructure/), [Dylan Patel](https://realalpha.app/s/ai-related-u-s-capex-could-approach-2-trillion-next-year/) ### Ten years and more · 8 - **Sergey Levine** (Co-founder, Physical Intelligence) said on Minus One on Oct 22 that robots will eventually generate vast amounts of physical-world data and learn faster by transferring skills across tasks. [Hear it at 21:08](https://realalpha.app/s/general-purpose-robots-can-bootstrap-their-own-data/) - **Max Tegmark** (Physics Professor, MIT) said on Into the Impossible With Brian Keating on Oct 23 that advanced AI could eventually reproduce itself and take control of its habitat from humans. [Hear it at 21:22](https://realalpha.app/s/superintelligent-ai-could-become-a-self-replicating-species/) - **Chamath Palihapitiya** (Founder & CEO, Social Capital) said on Moonshots on Oct 21 that AI growth will be constrained more by power shortages than by chips in the long run. [Hear it at 40:30](https://realalpha.app/s/power-shortages-will-constrain-ai-more-than-chips/) - **Bill Thompson** (Founder & CEO, Spartan Forge) said on The Joe Rogan Experience on Oct 21 that AI may eliminate many jobs within a decade, while making small businesses easier to start and increasing demand for human-made work. [Hear it at 2:37:53](https://realalpha.app/s/ai-could-make-local-production-and-human-made-work-more/) - And 4 more: [Alexander Wissner-Gross](https://realalpha.app/s/math-is-bulk-solved-first-because-it-is-easiest-to-verify/), [Stephen A. Schwarzman](https://realalpha.app/s/schwarzman-sees-ai-as-a-decades-long-investment-shift/), [Chris Dixon](https://realalpha.app/s/ai-agents-may-become-the-internets-most-numerous/), [George Kurtz](https://realalpha.app/s/enterprises-may-reserve-frontier-models-for-their-hardest/) ### Within five years · 7 - **Evan Conrad** (CEO, SF Compute) said on Upstarts on Oct 29 that GPU scarcity and rising prices will last at least four to five years, making compute a geopolitical lever. [Hear it at 34:11](https://realalpha.app/s/compute-financing-could-become-a-us-geopolitical-lever/) - **Alex Rives** (Head of Science, Biohub) said on The Cognitive Revolution on Oct 20 that AI models could help map cellular health and identify disease-linked cell states within five years. [Hear it at 45:58](https://realalpha.app/s/a-coordinated-effort-could-map-cellular-health-and-disease/) - **Andy Jassy** (CEO, Amazon) said on a16z Podcast on Oct 23 that inference will be used in most enterprise apps within five to seven years. [Hear it at 20:43](https://realalpha.app/s/ai-demand-is-concentrated-but-aws-wont-displace-core/) - **Matthew Prince** (Co-founder & CEO, Cloudflare) said on TBPN on Oct 22 that agents may generate a thousand times more traffic than people within five years, prompting micropayments and ads. [Hear it at 24:53](https://realalpha.app/s/agent-traffic-could-force-the-internet-to-charge-per-request/) - And 3 more: [Will Larson](https://realalpha.app/s/middle-managers-may-need-to-return-to-hands-on-engineering/), [David Friedberg](https://realalpha.app/s/governments-may-regulate-access-to-compute-rather-than-chips/), [Eric Schmidt](https://realalpha.app/s/ai-can-use-known-knowledge-but-novel-discovery-remains/) [Every answer to What will AI look like in the next few years? since October 2024 →](https://realalpha.app/q/ai-predictions/) ## [Are AI agents ready for real work?](https://realalpha.app/q/are-ai-agents-ready/) 27 of the 28 people who answered it in October say they’re ready for some work. 1 says they’re not ready yet. None say they’re ready now. ### Ready for some work · 27 - **Jake Stauch** (Co-founder & CEO, Serval) said on No Priors on Oct 29 that coding agents can build tools for other departments, with approvals and audit trails controlling what reaches production. [Hear it at 3:04](https://realalpha.app/s/it-can-bring-coding-agents-to-every-department/) - **Sundar Pichai** (CEO, Google and Alphabet) said on Sources on Oct 21 that agents can work across large codebases, while larger productivity gains still lie ahead. [Hear it at 8:30](https://realalpha.app/s/pichai-expects-a-productivity-leap-from-ai-agents-in-2027/) - **Bernardo Aceituno** (President, Stack AI) said on MTS on Oct 17 that agents complete five enterprise tasks at human-quality reliability, while governance and access control remain challenges. [Hear it at 1:02](https://realalpha.app/s/enterprise-agi-is-changing-the-challenge-from-capability-to/) - **Jad Naous** (CEO, Grepr) said on a16z Podcast on Oct 16 that agents can help deliver production features, but people still review and tie the work together. [Hear it at 17:52](https://realalpha.app/s/ai-cut-a-complex-data-lake-project-from-six-months-to-ten/) - And 23 more: [Han Wang](https://realalpha.app/s/agents-automate-code-but-still-face-business-decisions/), [Ian Quigley](https://realalpha.app/s/measured-chemical-landscapes-let-agents-run-thousands-of/), [Oege de Moor](https://realalpha.app/s/one-researcher-plus-ai-agents-found-a-chrome-full-chain/), [Jeanne Grosser](https://realalpha.app/s/vercel-runs-support-bdr-and-revops-with-agents/), [Tibo Sottiaux](https://realalpha.app/s/a-dot-warned-tibo-of-a-production-outage-before-a-live-demo/), [Erik Bernhardsson](https://realalpha.app/s/agents-write-most-of-modals-merged-code/), [Pierre Stock](https://realalpha.app/s/mistral-says-its-models-speed-senior-researchers-up-two-to/), [Arvind Krishna](https://realalpha.app/s/ai-written-code-can-outperform-when-agents-run-evals/), [Woodson Martin](https://realalpha.app/s/model-approval-can-stall-enterprise-ai-agents/), [Ian Weinberg](https://realalpha.app/s/local-agents-can-turn-a-prompt-into-a-blender-scene/), [Gavin Baker](https://realalpha.app/s/ai-agents-are-helping-investors-route-information-faster/), [Shaan Puri](https://realalpha.app/s/ai-assistants-can-help-with-tasks-people-avoid/), [Mark Cuban](https://realalpha.app/s/ai-agents-can-expand-output-and-create-new-work/), [Salim Ismail](https://realalpha.app/s/ai-agents-need-controls-as-well-as-autonomy/), [Eric Schmidt](https://realalpha.app/s/ai-agents-could-flatten-business-and-speed-up-work/), [Uday Kiran Medisetty](https://realalpha.app/s/production-replay-lets-agents-validate-fixes-against-real/), [Anjney Midha](https://realalpha.app/s/how-pooled-compute-lets-startups-and-public-users-share-gpu/), [Vlad Tenev](https://realalpha.app/s/agent-led-product-work-depends-on-automated-testing/), [Nilay Patel](https://realalpha.app/s/enterprise-users-tolerate-brittle-agents-consumers-walk-away/), [Dan Shipper](https://realalpha.app/s/why-every-personal-agent-at-every-died-and-one-company-agent/), [Yinon Costica](https://realalpha.app/s/wiz-atlas-makes-vulnerability-research-repeatable/), [Kevin Mandia](https://realalpha.app/s/cyber-agents-need-domain-expertise-and-layered-safety-checks/), [Sam Newman](https://realalpha.app/s/run-an-ai-software-factory-on-one-module-not-the-whole/) ### Not yet · 1 - **Diogo Almeida** (CEO, TypeSafe) said on The MAD Podcast on Oct 15 that polished, preference-tuned responses can make people overestimate reliability, while errors still limit automation. [Hear it at 27:12](https://realalpha.app/s/rlhf-made-models-pleasing-but-raised-automation-expectations/) [Every answer to Are AI agents ready for real work? since December 2024 →](https://realalpha.app/q/are-ai-agents-ready/) ## [Is AI dangerous?](https://realalpha.app/q/is-ai-dangerous/) 15 of the 24 people who answered it in October say it’s a serious risk. 9 say the risks are real but manageable. None say the fears are overblown. ### A serious risk · 15 - **Tudor Achim** (Co-founder & CEO, Harmonic) said on MTS on Oct 31 that AI-enabled attacks pose a severe cybersecurity risk to national infrastructure. [Hear it at 13:09](https://realalpha.app/s/critical-infrastructure-faces-bridge-scale-cybersecurity/) - **Tiago Sada** (CPO, World) said on MTS on Oct 20 that real-time deepfakes can cause serious, hard-to-reverse financial harm by manipulating markets or triggering fraudulent transfers. [Hear it at 1:16](https://realalpha.app/s/real-time-deepfakes-can-cause-lasting-financial-damage/) - **Nathan Lambert** (Founder, Trillium) said on a16z Podcast on Oct 14 that AI capabilities are already spreading while public cybersecurity is unprepared for them. [Hear it at 54:55](https://realalpha.app/s/scaling-laws-argue-against-a-fast-takeoff-but-cyber-risk-is/) - **Eric Ho** (Co-founder & CEO, Goodfire) said on The MAD Podcast on Oct 1 that agents can exploit weak sandboxes and vulnerabilities to escape containment and access outside organizations. [Hear it at 8:12](https://realalpha.app/s/impossible-tasks-and-weak-sandboxes-can-turn-agents-rogue/) - And 11 more: [Dan Primack](https://realalpha.app/s/compute-access-is-anthropics-practical-risk-while-safety/), [Max Tegmark](https://realalpha.app/s/superintelligent-ai-could-become-a-self-replicating-species/), [Ethan Elasky](https://realalpha.app/s/more-capable-models-can-produce-new-forms-of-misalignment/), [Emil Michael](https://realalpha.app/s/michael-objects-to-giving-ai-models-authority-over/), [Nate Soares](https://realalpha.app/s/automated-supply-chains-could-make-ai-self-sufficient/), [John Coogan](https://realalpha.app/s/locking-down-open-source-clashes-with-cryptos-libertarian/), [Sherry Turkle](https://realalpha.app/s/a-chatbot-affirmed-an-ice-agent-as-her-deportation-story/), [Peter Thiel](https://realalpha.app/s/thiel-sees-global-government-as-an-existential-risk-too/), [Nikesh Arora](https://realalpha.app/s/ai-vulnerability-findings-create-an-asymmetric-race/), [Eric Schmidt](https://realalpha.app/s/a-bot-incident-showed-why-ai-containment-matters/), [Garrison Lovely](https://realalpha.app/s/labor-replacement-connects-job-concerns-to-ai-safety/) ### Real but manageable · 9 - **Pierre Stock** (VP of Science, Mistral AI) said on The MAD Podcast on Oct 22 that a model’s attempted connection was contained, and monitoring and sandbox controls can address such software problems. [Hear it at 1:03:20](https://realalpha.app/s/a-model-pinged-its-trainer-from-inside-a-sandbox/) - **Nenad Tomašev** (Senior Staff Research Scientist, Google DeepMind) said on MTS on Oct 20 that unstructured agent swarms can cause harm, but norms and structure can make them governable. [Hear it at 21:43](https://realalpha.app/s/agent-swarms-need-structure-to-stay-manageable/) - **Jensen Huang** (Founder & CEO, NVIDIA) said on BG2 on Oct 15 that AI safety risks can be managed with engineering controls, testing, limited access, and monitoring. [Hear it at 30:39](https://realalpha.app/s/ai-safety-is-an-engineering-problem-containment-and-out-of/) - **Sam Altman** (CEO, OpenAI) said on Cheeky Pint on Oct 29 that iterative deployment lets people shape AI use, helping manage risks while capabilities reach users. [Hear it at 4:33](https://realalpha.app/s/openai-favors-iterative-deployment-to-keep-people-in-control/) - And 5 more: [Devon Thomas](https://realalpha.app/s/access-limits-can-reduce-misuse-but-cannot-eliminate-it/), [Salim Ismail](https://realalpha.app/s/ai-agents-need-controls-as-well-as-autonomy/), [Aghi Marietti](https://realalpha.app/s/ai-gateways-give-enterprises-a-checkpoint-to-stop-agent/), [Alan Chan](https://realalpha.app/s/why-an-ai-takeover-is-much-harder-than-the-rogue-agent-panic/), [Masayoshi Son](https://realalpha.app/s/son-argues-ai-safety-requires-progress-not-a-slowdown/) [Every answer to Is AI dangerous? since November 2024 →](https://realalpha.app/q/is-ai-dangerous/) ## [Which jobs will AI replace, and which are safe?](https://realalpha.app/q/which-jobs-will-ai-replace/) 15 of the 23 people who answered it in October say jobs will change more than vanish. 7 say AI will take over jobs. 1 says some work stays with people or grows. ### Jobs that change · 15 - **Marco Palladino** (CTO, Kong) said on MTS on Oct 20 that workers will shift from doing tasks to orchestrating agents. [Hear it at 19:49](https://realalpha.app/s/workers-will-shift-from-doing-tasks-to-orchestrating-agents/) - **Jad Naous** (CEO, Grepr) said on a16z Podcast on Oct 16 that engineers use agents to plan and implement work, then review and consolidate what they produce. [Hear it at 17:52](https://realalpha.app/s/ai-cut-a-complex-data-lake-project-from-six-months-to-ten/) - **Han Wang** (Co-founder, Mintlify) said on MTS on Oct 16 that agents take over writing and maintenance, while people steer content with intent and customer context. [Hear it at 13:15](https://realalpha.app/s/good-docs-need-structure-even-as-agents-take-over-writing/) - **Ali Ansari** (Founder & CEO, Micro1) said on Sabrina Halper Show on Oct 15 that experts take on new work as AI automates parts of their work. [Hear it at 22:36](https://realalpha.app/s/human-expertise-must-keep-updating-ai-models/) - And 11 more: [Jeanne Grosser](https://realalpha.app/s/vercel-runs-support-bdr-and-revops-with-agents/), [Erik Bernhardsson](https://realalpha.app/s/agents-write-most-of-modals-merged-code/), [Angela Jiang](https://realalpha.app/s/humans-own-system-design-for-claude-written-code/), [Guillermo Rauch](https://realalpha.app/s/engineers-should-manage-agents-not-blindly-ship-their-code/), [Mark Cuban](https://realalpha.app/s/personalized-drugs-could-make-treatment-fit-the-patient/), [Will Larson](https://realalpha.app/s/middle-managers-may-need-to-return-to-hands-on-engineering/), [Ethan Mollick](https://realalpha.app/s/ai-helps-students-learn-when-it-acts-as-a-tutor/), [Felix Rieseberg](https://realalpha.app/s/ai-shifts-managers-from-directing-work-to-enabling-it/), [Jay Madheswaran](https://realalpha.app/s/routine-legal-back-and-forth-may-move-between-agents/), [Prinz](https://realalpha.app/s/llms-help-lawyers-find-the-needle-but-leave-the-haystack/), [Uday Kiran Medisetty](https://realalpha.app/s/uber-still-requires-human-review-on-every-pr/) ### Jobs AI will replace · 7 - **Elon Musk** (CEO, SpaceX & Tesla; founder, xAI) said on The Builders with Walter Isaacson on Oct 20 that AI and robots could make necessary human work less important, leaving people to choose activities for meaning. [Hear it at 15:34](https://realalpha.app/s/robots-could-make-intelligence-far-more-economically/) - **Joe Rogan** (Host, The Joe Rogan Experience) said on The Joe Rogan Experience on Oct 31 that AI could disrupt jobs and leave people excluded from its gains. [Hear it at 2:03:55](https://realalpha.app/s/ai-wealth-needs-a-way-to-include-people-who-lose-work/) - **Bill Thompson** (Founder & CEO, Spartan Forge) said on The Joe Rogan Experience on Oct 21 that AI may eliminate many jobs, though abundant generated media could increase demand for live and handmade work. [Hear it at 2:37:53](https://realalpha.app/s/ai-could-make-local-production-and-human-made-work-more/) - **Eric Schmidt** (Former CEO, Google) said on In Good Company with Nicolai Tangen on Oct 21 that AI can eliminate coordination work and reduce the need for some planning and administrative roles. [Hear it at 35:56](https://realalpha.app/s/ai-could-remove-the-coordination-tax-in-large-companies/) - And 3 more: [Ed Elson](https://realalpha.app/s/microsofts-buyout-plan-could-take-headcount-below-pre/), [Nik Storonsky](https://realalpha.app/s/ai-is-shrinking-revoluts-teams-and-speeding-up-work/), [Garrison Lovely](https://realalpha.app/s/ai-exposure-is-narrowing-entry-level-hiring/) ### Jobs that are safe or grow · 1 - **David Friedberg** (CEO, Ohalo) said on All-In on Oct 10 that truck drivers are not being replaced because their work depends on the physical world. [Hear it at 39:21](https://realalpha.app/s/math-proofs-fall-fastest-where-the-whole-test-loop-runs-in/) [Every answer to Which jobs will AI replace, and which are safe? since October 2024 →](https://realalpha.app/q/which-jobs-will-ai-replace/) ## [Is AI paying off for companies?](https://realalpha.app/q/is-ai-paying-off/) 10 of the 19 people who answered it in October say it’s paying off. 8 say it pays off in places. 1 says it isn’t paying off yet. ### It’s paying off · 10 - **Katelyn Lesse** (Head of Engineering, Anthropic) said on Latent Space on Oct 27 that claude helped rewrite the API, cutting CPU use by 40–50% and improving response speed. [Hear it at 56:02](https://realalpha.app/s/claude-helped-cut-messages-api-cpu-use-by-up-to-50/) - **Jad Naous** (CEO, Grepr) said on a16z Podcast on Oct 16 that AI can deliver production work much faster, turning a six-month data-lake feature into a ten-day project. [Hear it at 17:52](https://realalpha.app/s/ai-cut-a-complex-data-lake-project-from-six-months-to-ten/) - **Jason Ma** (Co-founder & Chief Scientist, Dyna Robotics) said on Latent Space on Oct 13 that companies see enough value from some AI robots to expand deployments across many sites. [Hear it at 38:50](https://realalpha.app/s/multi-site-rollouts-not-pilots-prove-a-robot-creates-value/) - **Jeanne Grosser** (COO, Vercel) said on Grit on Oct 5 that companies use agents to resolve most support cases, answer internal questions, and increase sales output without adding reps. [Hear it at 57:49](https://realalpha.app/s/vercel-runs-support-bdr-and-revops-with-agents/) - And 6 more: [Pierre Stock](https://realalpha.app/s/mistral-says-its-models-speed-senior-researchers-up-two-to/), [Arvind Krishna](https://realalpha.app/s/ai-written-code-can-outperform-when-agents-run-evals/), [Dylan Patel](https://realalpha.app/s/ai-revenue-may-justify-infrastructure-spending-sooner-than/), [David Friedberg](https://realalpha.app/s/ai-customers-may-earn-multiples-of-their-model-spend/), [Nik Storonsky](https://realalpha.app/s/ai-is-shrinking-revoluts-teams-and-speeding-up-work/), [Meerah Rajavel](https://realalpha.app/s/palo-alto-uses-a-harness-to-measure-ai-assisted-development/) ### Pays off in places · 8 - **Alex Danilowicz** (CEO, Magic Patterns) said on Lightcone on Oct 22 that some AI design workflows save hours when teams provide company context upfront. [Hear it at 14:33](https://realalpha.app/s/bring-company-context-into-ai-design-before-prompting/) - **Ethan Mollick** (Professor, Wharton School) said on Big Technology Podcast on Oct 20 that companies can get large savings when they redesign work around AI. [Hear it at 13:21](https://realalpha.app/s/ai-gains-require-companies-to-redesign-how-teams-work/) - **Cal Newport** (Professor of Computer Science, Georgetown University) said on Prof G Markets on Oct 9 that companies are investing heavily, but returns are mixed and much is still experimental. [Hear it at 55:33](https://realalpha.app/s/enterprise-ai-adoption-is-real-but-harder-to-turn-into-roi/) - **Nilay Patel** (Editor-in-chief, The Verge) said on Decoder on Oct 8 that businesses keep retrying brittle agents because the potential savings justify the effort. [Hear it at 16:04](https://realalpha.app/s/enterprise-users-tolerate-brittle-agents-consumers-walk-away/) - And 4 more: [Dave Blundin](https://realalpha.app/s/ai-is-cutting-the-cost-of-replacing-legacy-software/), [Uday Kiran Medisetty](https://realalpha.app/s/uber-is-shifting-from-diff-counts-to-feature-velocity/), [Rodney Zemmel](https://realalpha.app/s/ai-implementation-needs-business-and-technology-leaders/), [Sean Frank](https://realalpha.app/s/ridge-generates-quarterly-reports-with-three-ai-prompts/) ### Not yet · 1 - **Chamath Palihapitiya** (Founder & CEO, Social Capital) said on All-In on Oct 23 that broader economic value remains unproven while spending is concentrated in unprofitable startups. [Hear it at 1:49](https://realalpha.app/s/a-small-group-of-startups-drives-most-ai-spending/) [Every answer to Is AI paying off for companies? since December 2024 →](https://realalpha.app/q/is-ai-paying-off/) ## [How do AI companies make money?](https://realalpha.app/q/how-do-ai-companies-make-money/) 6 of the 13 people who answered it in October say the model is changing. 4 say margins are hard. 3 say it’s a good business. ### The model is changing · 6 - **Misha Laskin** (Co-Founder & CEO, ReflectionAI) said on No Priors on Oct 9 that open-model businesses can earn through services and paid inference tools built around the weights. [Hear it at 22:29](https://realalpha.app/s/enterprises-rent-intelligence-first-then-want-to-own-it/) - **Nilay Patel** (Editor-in-chief, The Verge) said on Decoder on Oct 8 that AI companies are trying different models, including transaction cuts and enterprise revenue, while some subscriptions lose money. [Hear it at 33:20](https://realalpha.app/s/meta-can-fund-free-agents-with-ads-while-openai-cannot/) - **Daron Acemoğlu** (Economist, MIT) said on Silicon Valley Girl on Oct 6 that the money is shifting toward data and applications, while open source and diffusion limit what foundation models can earn. [Hear it at 40:08](https://realalpha.app/s/the-money-in-ai-is-in-data-and-applications-not-models/) - **Rory O'Driscoll** (General Partner, Scale) said on 20VC on Oct 22 that independent inference providers can grow by serving open models at lower token costs, capturing spend beyond model makers. [Hear it at 31:21](https://realalpha.app/s/inference-providers-may-capture-spend-beyond-model-makers/) - And 2 more: [Vlad Tenev](https://realalpha.app/s/robinhood-plans-to-monetize-agent-tools-through/), [Sean Frank](https://realalpha.app/s/ai-shopping-agents-may-take-fees-from-both-sides/) ### Margins are hard · 4 - **Erik Bernhardsson** (Co-founder & CEO, Modal) said on Unsupervised Learning on Oct 28 that basic token serving becomes a race to the bottom, while broader platforms can offer more durable differentiation. [Hear it at 21:21](https://realalpha.app/s/inference-providers-need-a-platform-beyond-token-serving/) - **Steve Eisman** (Investor) said on Prof G Markets on Oct 28 that future compute commitments raise expenses, and price cuts could pressure earnings unless revenue grows. [Hear it at 5:42](https://realalpha.app/s/anthropics-commitments-raise-costs-while-price-cuts-test-its/) - **Paul Kedrosky** (Managing Partner, SK Ventures) said on Prof G Markets on Oct 27 that massive compute commitments and infrastructure spending make future margins hard to judge. [Hear it at 8:16](https://realalpha.app/s/anthropics-future-compute-costs-dwarf-its-current-losses/) - **Garrison Lovely** (Freelance journalist & author) said on Future of Life Institute Podcast on Oct 13 that training frontier models costs so much that companies reinvest product profits, betting revenue will eventually outgrow costs. [Hear it at 32:17](https://realalpha.app/s/ai-training-costs-matter-more-than-serving-customers/) ### A good business · 3 - **David Sacks** (Co-founder & partner, Craft Ventures) said on All-In on Oct 30 that this AI company is capital-efficient, with high and improving gross margins. [Hear it at 6:40](https://realalpha.app/s/anthropics-reported-80b-loss-was-not-cash-burn/) - **Peter Diamandis** (Founder, XPRIZE) said on Moonshots on Oct 28 that Anthropic earns more than two dollars in revenue per dollar of compute, though it faces capacity and customer-concentration risks. [Hear it at 31:35](https://realalpha.app/s/anthropic-now-earns-2-08-per-compute-dollar/) - **Dylan Patel** (Founder & CEO, SemiAnalysis) said on Big Technology Podcast on Oct 9 that AI labs can have high inference margins, low customer acquisition costs, and R&D growing slower than revenue. [Hear it at 29:20](https://realalpha.app/s/anthropic-pairs-high-inference-margins-with-low-acquisition/) [Every answer to How do AI companies make money? since February 2025 →](https://realalpha.app/q/how-do-ai-companies-make-money/) ## [Is there still a GPU shortage?](https://realalpha.app/q/is-there-a-gpu-shortage/) 10 of the 12 people who answered it in October say compute is still short. 2 say the shortage is easing. None say a glut is coming. ### Still short · 10 - **Sundar Pichai** (CEO, Google) said on Big Technology Podcast on Oct 20 that frontier AI compute remains a priority, and present shortages reflect data-center decisions made years earlier. [Hear it at 15:04](https://realalpha.app/s/google-puts-frontier-ai-compute-ahead-of-other-uses/) - **Matt Garman** (CEO, AWS) said on a16z Podcast on Oct 8 that demand still exceeds available capacity, so some requests cannot be fulfilled. [Hear it at 17:04](https://realalpha.app/s/aws-reserves-gpu-capacity-to-support-startups/) - **Evan Conrad** (CEO, SF Compute) said on Upstarts on Oct 29 that GPU scarcity is expected to last at least four to five years. [Hear it at 34:11](https://realalpha.app/s/compute-financing-could-become-a-us-geopolitical-lever/) - **Satya Nadella** (CEO, Microsoft) said on 20VC on Oct 26 that silicon capacity remains a long-lead constraint that cannot be brought online quickly. [Hear it at 15:47](https://realalpha.app/s/ai-training-budgets-ultimately-depend-on-gdp-growth/) - And 6 more: [Gavin Baker](https://realalpha.app/s/gpu-supply-is-tight-despite-the-capacity-coming-online/), [Dan Primack](https://realalpha.app/s/compute-access-is-anthropics-practical-risk-while-safety/), [Andy Jassy](https://realalpha.app/s/customer-demand-guides-amazons-ai-infrastructure-spending/), [Jordan Nanos](https://realalpha.app/s/neocloud-contracts-differ-from-hyperscalers-self-build-plans/), [Nikesh Arora](https://realalpha.app/s/enterprises-favor-open-models-but-still-need-hyperscalers/), [Jake Cooper](https://realalpha.app/s/agent-growth-could-make-compute-the-next-infrastructure/) ### Easing · 2 - **Chamath Palihapitiya** (Founder & CEO, Social Capital) said on Moonshots on Oct 21 that power, rather than chips, is the main constraint on expanding AI capacity. [Hear it at 40:30](https://realalpha.app/s/power-shortages-will-constrain-ai-more-than-chips/) - **Scott Galloway** said on Prof G Markets on Oct 26 that the GPU bottleneck has eased, with power now limiting data-center capacity. [Hear it at 51:03](https://realalpha.app/s/ai-chip-delays-may-reflect-power-bottlenecks-not-weak-demand/) [Every answer to Is there still a GPU shortage? since April 2025 →](https://realalpha.app/q/is-there-a-gpu-shortage/) ## [Is AI a bubble?](https://realalpha.app/q/is-ai-a-bubble/) 7 of the 12 people who answered it in October say it isn't a bubble. 5 say it partly is. None say it is a bubble. ### Not a bubble · 7 - **Sundar Pichai** (CEO, Google and Alphabet) said on Sources on Oct 21 that demand and committed contracts support continued AI investment, though volatility still calls for discipline. [Hear it at 22:47](https://realalpha.app/s/google-justifies-ai-spending-with-demand-and-capability/) - **Jensen Huang** (Founder & CEO, NVIDIA) said on BG2 on Oct 15 that AI compute spending is backed by demand from classical computing, frontier labs, and thousands of startups. [Hear it at 1:01:57](https://realalpha.app/s/nvidia-demand-rests-on-three-segments-not-two-ai-labs/) - **Mark Zuckerberg** (Founder & CEO, Meta) said on Cheeky Pint on Oct 12 that their compute spending is backed by demand from other buyers if AI succeeds overall. [Hear it at 56:35](https://realalpha.app/s/why-metas-compute-bet-is-less-risky-than-the-100b-headline/) - **Matthew Harris** (Partner, Bain Capital Ventures) said on How I Invest on Oct 26 that continued fast growth could make the reported valuation reasonable. [Hear it at 32:44](https://realalpha.app/s/anthropics-growth-could-support-its-reported-2-trillion/) - And 3 more: [Andy Jassy](https://realalpha.app/s/customer-demand-guides-amazons-ai-infrastructure-spending/), [Dylan Patel](https://realalpha.app/s/ai-revenue-may-justify-infrastructure-spending-sooner-than/), [Winfield Sickles](https://realalpha.app/s/cloud-revenue-is-beginning-to-reflect-ai-infrastructure/) ### Partly a bubble · 5 - **Ed Elson** said on Prof G Markets on Oct 26 that private AI markets contain an unpriced bubble, while public markets price the risks more cautiously. [Hear it at 55:55](https://realalpha.app/s/public-markets-price-ai-risk-more-cautiously-than-private/) - **Mark Cuban** (Entrepreneur, Cost Plus Drugs) said on The Diary Of A CEO on Oct 26 that hyperscalers may create a debt bubble by borrowing heavily for AI infrastructure, especially if rates rise and credit tightens. [Hear it at 51:34](https://realalpha.app/s/ai-infrastructure-debt-could-intensify-a-market-downturn/) - **Steve Eisman** said on The Real Eisman Playbook on Oct 16 that some AI infrastructure spending looks fragile because debt is high and revenue claims may not prove real demand. [Hear it at 4:34](https://realalpha.app/s/ai-data-center-debt-is-testing-confidence-in-the-boom/) - **Robert Armstrong** (US Financial Commentator, Financial Times) said on Prof G Markets on Oct 27 that the AI infrastructure boom will be followed by restructuring and consolidation. [Hear it at 27:03](https://realalpha.app/s/ai-infrastructure-financing-is-headed-for-consolidation/) - And 1 more: [Scott Galloway](https://realalpha.app/s/ai-chip-delays-may-reflect-power-bottlenecks-not-weak-demand/) [Every answer to Is AI a bubble? since May 2025 →](https://realalpha.app/q/is-ai-a-bubble/) ## [Should AI be regulated?](https://realalpha.app/q/should-ai-be-regulated/) 9 of the 11 people who answered it in October say light, targeted rules. 2 say regulate it now. None say hold off. ### Light, targeted rules · 9 - **Richard Socher** (Co-founder & CEO, Recursive) said on MTS on Oct 26 that rules should target high-stakes AI applications, not intelligence itself, because banning self-improvement would require invasive surveillance. [Hear it at 11:05](https://realalpha.app/s/banning-self-improvement-would-require-invasive-surveillance/) - **Ali Ansari** (Founder & CEO, Micro1) said on Sabrina Halper Show on Oct 15 that valuable training data should not be exported to geopolitical rivals because it could help them catch up. [Hear it at 49:24](https://realalpha.app/s/training-data-is-a-strategic-asset-that-should-not-be/) - **Max Tegmark** (Physics Professor, MIT) said on Into the Impossible With Brian Keating on Oct 23 that government should be able to compel AI labs to disclose information and undergo independent scrutiny. [Hear it at 11:24](https://realalpha.app/s/ai-labs-can-keep-the-public-in-the-dark/) - **Eric Schmidt** (Former CEO, Google) said on In Good Company with Nicolai Tangen on Oct 21 that lighter rules would help European AI companies build and compete. [Hear it at 17:41](https://realalpha.app/s/europe-needs-to-make-it-easier-to-build-ai-companies/) - And 5 more: [Alan Chan](https://realalpha.app/s/pace-ai-progress-with-concrete-pause-triggers-not-a-growth/), [John Coogan](https://realalpha.app/s/locking-down-open-source-clashes-with-cryptos-libertarian/), [Ali Partovi](https://realalpha.app/s/ai-labs-could-face-liability-when-their-models-choose/), [Chamath Palihapitiya](https://realalpha.app/s/ai-oversight-should-inspect-labs-founding-documents/), [Reid Hoffman](https://realalpha.app/s/ai-trust-needs-claims-that-can-be-audited/) ### Regulate it now · 2 - **Garrison Lovely** (Freelance journalist & author) said on Future of Life Institute Podcast on Oct 13 that development should require safety and informed public consent. [Hear it at 8:04](https://realalpha.app/s/replacing-workers-can-weaken-democratic-accountability/) - **Nate Soares** (President, Machine Intelligence Research Institute) said on Odd Lots on Oct 12 that governments should publicly estimate catastrophic AI risks and charge developers for taking them. [Hear it at 45:02](https://realalpha.app/s/public-actuaries-could-put-a-price-on-catastrophic-ai-risk/) [Every answer to Should AI be regulated? since February 2025 →](https://realalpha.app/q/should-ai-be-regulated/) ## [Will AI make cyberattacks worse?](https://realalpha.app/q/will-ai-make-cyberattacks-worse/) 5 of the 10 people who answered it in October say attackers gain more. 3 say defenders gain more. 2 say both sides gain. ### Attackers gain · 5 - **Tudor Achim** (Co-founder & CEO, Harmonic) said on MTS on Oct 31 that AI makes attacks on vulnerable infrastructure more feasible, creating a severe risk to national infrastructure. [Hear it at 13:09](https://realalpha.app/s/critical-infrastructure-faces-bridge-scale-cybersecurity/) - **George Kurtz** (President & CEO, CrowdStrike) said on No Priors on Oct 22 that AI increases vulnerability volume and speeds exploitation, raising the pace defenders must handle. [Hear it at 30:01](https://realalpha.app/s/ai-attacks-raise-the-pace-not-create-a-new-attack-class/) - **Dylan Patel** (Founder & CEO, SemiAnalysis) said on Big Technology Podcast on Oct 9 that open-source models can hack some neoclouds. [Hear it at 39:34](https://realalpha.app/s/a-neocloud-security-gap-exposed-other-customers-workloads/) - **Kevin Mandia** (Founder & CEO, Armadin) said on a16z Podcast on Oct 6 that machine-speed attacks can outrun human-led response, so defenses need AI to detect and respond quickly. [Hear it at 21:09](https://realalpha.app/s/ai-has-to-take-over-the-detect-and-respond-loop/) - And 1 more: [Matthew Prince](https://realalpha.app/s/ai-helps-attackers-navigate-systems-after-a-breach/) ### Defenders gain · 3 - **Devon Thomas** (CEO, Abliteration) said on MTS on Oct 17 that defenders can use unrestricted models to test systems before attackers do, while models mostly recombine known techniques. [Hear it at 3:05](https://realalpha.app/s/defenders-need-to-test-systems-before-attackers-do/) - **Greg Brockman** (President & Co-founder, OpenAI) said on No Priors on Oct 15 that AI gives defenders a temporary lead to find and fix flaws before capabilities spread. [Hear it at 26:56](https://realalpha.app/s/ai-gives-defenders-a-window-to-find-and-fix-security-flaws/) - **Yinon Costica** (Co-founder, Wiz) said on Sabrina Halper Show on Oct 7 that defenders can use AI and their internal context to find weaknesses and respond faster than attackers. [Hear it at 38:32](https://realalpha.app/s/ai-can-give-defenders-a-compounding-advantage/) ### Both sides gain · 2 - **Guillermo Rauch** (Founder & CEO, Vercel) said on Sourcery on Oct 16 that AI strengthens both offensive and defensive security efforts, creating a contest between them. [Hear it at 47:45](https://realalpha.app/s/a-public-challenge-helped-test-vercels-agent-sandbox/) - **Nikesh Arora** (CEO, Palo Alto Networks) said on All-In on Oct 23 that AI can help uncover hard-to-find vulnerabilities, while released patches can also help attackers exploit unpatched systems. [Hear it at 16:25](https://realalpha.app/s/ai-is-exposing-old-software-debt-faster-than-teams-can-patch/) [Every answer to Will AI make cyberattacks worse? since January 2025 →](https://realalpha.app/q/will-ai-make-cyberattacks-worse/) ## [Is there enough power for AI?](https://realalpha.app/q/is-there-enough-power-for-ai/) 8 of the 9 people who answered it in October say power is the bottleneck. 1 says it’s tight but solvable. None say power isn’t the problem. ### Power is the bottleneck · 8 - **Aravind Srinivas** (Founder & CEO, Perplexity) said on Sourcery on Oct 12 that power, permits, and interconnection queues are bottlenecking data center build-outs. [Hear it at 29:26](https://realalpha.app/s/long-context-agents-make-local-hardware-cost-competitive/) - **Satya Nadella** (CEO, Microsoft) said on 20VC on Oct 26 that power is a long-lead constraint that cannot be brought online quickly. [Hear it at 15:47](https://realalpha.app/s/ai-training-budgets-ultimately-depend-on-gdp-growth/) - **Misha Laskin** (CEO & Co-founder, Reflection) said on Sources on Oct 5 that land, power, and data center shells can bottleneck compute deployment. [Hear it at 49:10](https://realalpha.app/s/data-center-owners-need-open-models-and-a-software-partner/) - **Chamath Palihapitiya** (Founder & CEO, Social Capital) said on Moonshots on Oct 21 that electricity supply and local permitting will constrain AI data-center expansion. [Hear it at 40:30](https://realalpha.app/s/power-shortages-will-constrain-ai-more-than-chips/) - And 4 more: [Pat Gelsinger](https://realalpha.app/s/energy-capacity-will-cap-the-ai-buildout/), [Scott Galloway](https://realalpha.app/s/ai-chip-delays-may-reflect-power-bottlenecks-not-weak-demand/), [Ethan Mollick](https://realalpha.app/s/mollick-sees-no-clear-technical-wall-to-steady-ai-gains/), [Nick Del Deo](https://realalpha.app/s/power-access-favors-experienced-data-center-operators/) ### Tight but solvable · 1 - **Baiju Bhatt** (Founder & CEO, Cowboy Space) said on The Shawn Ryan Show on Oct 27 that orbiting solar arrays could supply data centers with substantial power without using local grids or land. [Hear it at 10:45](https://realalpha.app/s/orbit-could-ease-data-centers-energy-and-land-demands/) [Every answer to Is there enough power for AI? since April 2025 →](https://realalpha.app/q/is-there-enough-power-for-ai/) ## [Is SaaS dead?](https://realalpha.app/q/is-saas-dead/) 6 of the 8 people who answered it in October say SaaS will change. 2 say SaaS survives. None say SaaS is dying. ### SaaS changes · 6 - **Ari Weinstein** (Head of Computer Use, OpenAI) said on a16z Podcast on Oct 26 that legacy software will remain, while agents work with it and new tailored applications emerge. [Hear it at 25:23](https://realalpha.app/s/agents-will-extend-legacy-software-and-enable-new-products/) - **Jake Stauch** (Co-founder & CEO, Serval) said on No Priors on Oct 29 that AI may consolidate point solutions into core platforms while systems-of-record categories persist. [Hear it at 23:17](https://realalpha.app/s/ai-may-shrink-saas-sprawl-without-replacing-systems-of/) - **Guillermo Rauch** (Founder & CEO, Vercel) said on Sourcery on Oct 16 that AI enables personalized software, while systems of record remain important as data sources for agents. [Hear it at 24:13](https://realalpha.app/s/coding-agents-make-personalized-software-viable/) - **Nikesh Arora** (Chairman & CEO, Palo Alto Networks) said on All-In on Oct 21 that enterprises may consolidate many software vendors into a few core platforms, though highly reliable systems may resist replacement. [Hear it at 35:19](https://realalpha.app/s/ai-could-consolidate-hundreds-of-enterprise-software-vendors/) - And 2 more: [David Sacks](https://realalpha.app/s/ai-can-lower-software-prices-while-expanding-software-demand/), [Salim Ismail](https://realalpha.app/s/ai-models-are-becoming-capable-enough-to-use-existing/) ### SaaS survives · 2 - **Dylan Field** (Co-founder & CEO, Figma) said on All-In on Oct 19 that saas companies can still grow by building focused products around users and professional workflows. [Hear it at 14:08](https://realalpha.app/s/model-progress-does-not-automatically-create-a-good-product/) - **Matt Hartman** (Founder & Managing Partner, Factorial Capital) said on AI & I on Oct 14 that saas still sells because buyers pay for ongoing maintenance and upkeep. [Hear it at 35:23](https://realalpha.app/s/vibe-coding-his-own-app-taught-an-investor-why-saas-still/) [Every answer to Is SaaS dead? since March 2025 →](https://realalpha.app/q/is-saas-dead/) ## [Is open-source AI catching up?](https://realalpha.app/q/is-open-source-ai-catching-up/) 6 of the 7 people who answered it in October say open models are close behind. 1 says they’re falling behind. None say they’re behind but useful. ### Close behind · 6 - **Arvind Krishna** (Chairman & CEO, IBM) said on All-In on Oct 21 that open-weight models can handle most enterprise work at much lower cost, while frontier models remain for high-value tasks. [Hear it at 9:01](https://realalpha.app/s/open-weight-models-could-handle-most-enterprise-ai-at-lower/) - **Guillermo Rauch** (Founder & CEO, Vercel) said on Sourcery on Oct 16 that open models are good enough for many routine tasks and handle most of Vercel’s token traffic. [Hear it at 52:04](https://realalpha.app/s/open-models-now-handle-most-token-traffic-on-vercel/) - **Marjorie Janiewicz** said on MTS on Oct 30 that open models can match closed-source frontier models without billions, using efficient design. [Hear it at 11:40](https://realalpha.app/s/efficient-task-specific-models-can-compete-on-less-funding/) - **Alexander Wissner-Gross** (Founder, Reified) said on Moonshots on Oct 28 that open-weight models are only weeks behind the U.S. frontier and put pressure on closed labs. [Hear it at 54:32](https://realalpha.app/s/kimis-open-weights-are-closing-the-frontier-gap/) - And 2 more: [John Coogan](https://realalpha.app/s/open-source-models-trail-closed-ones-by-about-six-months/), [George Kurtz](https://realalpha.app/s/enterprises-may-reserve-frontier-models-for-their-hardest/) ### Falling behind · 1 - **Dylan Patel** (Founder & CEO, SemiAnalysis) said on Big Technology Podcast on Oct 9 that frontier labs keep stronger models private to preserve their lead over open source. [Hear it at 54:52](https://realalpha.app/s/frontier-labs-are-holding-back-models-before-public-release/) [Every answer to Is open-source AI catching up? since March 2025 →](https://realalpha.app/q/is-open-source-ai-catching-up/) ## [Will AI replace software engineers?](https://realalpha.app/q/will-ai-replace-software-engineers/) All 5 people who answered it in October say the job changes. None say AI will replace engineers. None say we will need more engineers. ### The job changes · 5 - **Jad Naous** (CEO, Grepr) said on a16z Podcast on Oct 16 that engineers use agents to plan and implement work, then review and consolidate what they produce. [Hear it at 17:52](https://realalpha.app/s/ai-cut-a-complex-data-lake-project-from-six-months-to-ten/) - **Erik Bernhardsson** (Co-founder & CEO, Modal) said on Unsupervised Learning on Oct 28 that agents write much of the code, while engineers use them and still oversee reliability and scale. [Hear it at 27:01](https://realalpha.app/s/agents-write-most-of-modals-merged-code/) - **Angela Jiang** (Head of Product, Anthropic) said on Latent Space on Oct 27 that software engineers shift toward system design and quality while Claude writes much of the implementation. [Hear it at 1:20:59](https://realalpha.app/s/humans-own-system-design-for-claude-written-code/) - **Guillermo Rauch** (Founder & CEO, Vercel) said on Sourcery on Oct 16 that engineers increasingly manage agents, review their code, and check that their outputs are safe. [Hear it at 11:32](https://realalpha.app/s/engineers-should-manage-agents-not-blindly-ship-their-code/) - And 1 more: [Uday Kiran Medisetty](https://realalpha.app/s/uber-still-requires-human-review-on-every-pr/) [Every answer to Will AI replace software engineers? since February 2025 →](https://realalpha.app/q/will-ai-replace-software-engineers/) ## [Is AI progress slowing down?](https://realalpha.app/q/is-ai-progress-slowing/) 2 of the 4 people who answered it in October say it’s shifting to new methods. 1 says progress is slowing. 1 says progress is speeding up. ### Shifting to new methods · 2 - **Misha Laskin** (Co-Founder & CEO, ReflectionAI) said on No Priors on Oct 9 that progress is shifting toward reinforcement learning and AI-assisted research as ways to improve efficiency. [Hear it at 8:50](https://realalpha.app/s/frontier-training-costs-rise-4x-per-generation-and-ai-speeds/) - **Ethan Mollick** (Professor, Wharton School) said on Big Technology Podcast on Oct 20 that reasoning and agent swarms are overcoming barriers, and AI ability continues making steady gains. [Hear it at 31:08](https://realalpha.app/s/mollick-sees-no-clear-technical-wall-to-steady-ai-gains/) ### Slowing down · 1 - **Dylan Field** (Co-founder & CEO, Figma) said on All-In on Oct 19 that creative AI progresses more slowly because its results are difficult to verify. [Hear it at 28:49](https://realalpha.app/s/creative-ai-progresses-slowly-because-its-results-are-hard/) ### Speeding up · 1 - **Garrison Lovely** (Freelance journalist & author) said on Future of Life Institute Podcast on Oct 13 that measured AI task horizons are doubling every three to seven months. [Hear it at 20:40](https://realalpha.app/s/ai-task-horizons-have-doubled-every-three-to-seven-months/) [Every answer to Is AI progress slowing down? since November 2024 →](https://realalpha.app/q/is-ai-progress-slowing/) ## [Will AI boost economic growth?](https://realalpha.app/q/will-ai-boost-economic-growth/) 3 of the 4 people who answered it in October say a big boost. 1 says a modest boost. None say little so far. ### A big boost · 3 - **Elon Musk** (CEO, SpaceX & Tesla; founder, xAI) said on The Builders with Walter Isaacson on Oct 20 that digital AI could raise goods and services output by about 20–30%, with robots making intelligence more economically valuable. [Hear it at 15:34](https://realalpha.app/s/robots-could-make-intelligence-far-more-economically/) - **Prinz** said on MTS on Oct 25 that AI-driven scientific discoveries could produce major economic growth, though the scale and timing are uncertain. [Hear it at 21:06](https://realalpha.app/s/scientific-discovery-could-drive-unusually-fast-growth/) - **Dylan Patel** (Founder & CEO, SemiAnalysis) said on Big Technology Podcast on Oct 9 that AI could help the economy grow fast enough to keep labor income from shrinking. [Hear it at 14:36](https://realalpha.app/s/ai-could-shift-value-to-capital-without-shrinking-labor/) ### A modest boost · 1 - **Peter Thiel** (Co-founder, PayPal) said on The Builders with Walter Isaacson on Oct 22 that AI can raise per-person productivity and support growth better than globalization. [Hear it at 15:36](https://realalpha.app/s/thiel-sees-ai-as-a-better-growth-engine-than-globalization/) [Every answer to Will AI boost economic growth? since June 2025 →](https://realalpha.app/q/will-ai-boost-economic-growth/) ## [How does OpenAI make money, and is it profitable?](https://realalpha.app/q/is-openai-profitable/) 2 of the 3 people who answered it in October doubt it pays off. 1 says how it earns. None say it will pay off. ### Doubts · 2 - **Nilay Patel** (Editor-in-chief, The Verge) said on Decoder on Oct 8 that OpenAI relies on enterprise revenue and likely loses money on $100-a-month users. [Hear it at 33:20](https://realalpha.app/s/meta-can-fund-free-agents-with-ads-while-openai-cannot/) - **Ed Elson** said on Prof G Markets on Oct 8 that its research progress does not prove profitability, and its losses are expanding. [Hear it at 32:06](https://realalpha.app/s/openais-research-progress-does-not-prove-its-business-model/) ### How OpenAI makes money · 1 - **Matthew Harris** (Partner, Bain Capital Ventures) said on How I Invest on Oct 26 that they expect to earn by selling their own model tokens through the coding platform. [Hear it at 27:59](https://realalpha.app/s/openais-open-codex-strategy-could-win-the-application-layer/) [Every answer to How does OpenAI make money, and is it profitable? since August 2025 →](https://realalpha.app/q/is-openai-profitable/) ## [Is China catching up in AI?](https://realalpha.app/q/is-china-catching-up-in-ai/) All 3 people who answered it in October say China is close behind. None say China is level or ahead. None say China is well behind. ### Close behind · 3 - **Nathan Lambert** (Founder, Trillium) said on a16z Podcast on Oct 14 that chinese labs match U.S. talent but lack compute, and distillation only partly narrows the gap. [Hear it at 8:53](https://realalpha.app/s/why-distillation-explains-less-of-chinas-ai-progress-than/) - **Alexander Wissner-Gross** (Founder, Reified) said on Moonshots on Oct 28 that China’s AI frontier is only weeks behind the U.S. public frontier and may have passed it. [Hear it at 54:32](https://realalpha.app/s/kimis-open-weights-are-closing-the-frontier-gap/) - **Eric Schmidt** (Former CEO, Google) said on In Good Company with Nicolai Tangen on Oct 21 that Chinese models trail by about six months, but free access could help them gain users. [Hear it at 24:03](https://realalpha.app/s/free-chinese-models-could-win-users-despite-trailing-by/) [Every answer to Is China catching up in AI? since December 2024 →](https://realalpha.app/q/is-china-catching-up-in-ai/) ## [Is vibe coding the future?](https://realalpha.app/q/is-vibe-coding-the-future/) All 3 people who answered it in October say it’s for some software. None say it’s the future. None say it’s not for real products. ### For some software · 3 - **Matt Hartman** (Founder & Managing Partner, Factorial Capital) said on AI & I on Oct 14 that vibe coding can build personal apps, but those apps still need ongoing maintenance. [Hear it at 35:23](https://realalpha.app/s/vibe-coding-his-own-app-taught-an-investor-why-saas-still/) - **Brian Kernighan** (Member of Technical Staff, Bell Labs) said on The Peterman Pod on Oct 26 that AI-generated code can help, but users need expertise to catch errors. [Hear it at 1:11:12](https://realalpha.app/s/ai-generated-code-still-needs-expertise-to-verify/) - **Sam Newman** (Independent consultant & author) said on The Pragmatic Engineer on Oct 7 that AI can build some low-risk modules without code review, but people should stay close to critical parts. [Hear it at 1:44:55](https://realalpha.app/s/run-an-ai-software-factory-on-one-module-not-the-whole/) [Every answer to Is vibe coding the future? since March 2025 →](https://realalpha.app/q/is-vibe-coding-the-future/) How we count. Every answer is a moment of a long interview that came out in October 2026, cut by Real Alpha and filed under how it answers by a model that read its claim and quote. The words after “said that” are that reading; the speaker’s exact words play from the link. A question is here when at least three people answered it that month, and each person counts once, on the side of their newest answer then. Each question’s page has every answer since we began.