No Priors Ep. 144 | The 2026 AI Forecast with Sarah & Elad

2025-12-19 - 41 min - source - Read full transcript

Sarah Guo (host)Elad Gil (host)AaronBen SpectreAsh Spectre

Key insights

AI adoption is fastest among the professions that are traditionally the slowest to adopt new technology.

Sarah cites a report on physician AI adoption (documentation and clinical decision support via tools like Abridge and Open Evidence) showing doctors, lawyers, and certain accounting types adopting AI unusually fast despite being historically conservative technology adopters. She calls this pattern real but underdiscussed.

ai-adoption-and-verticals

2026 will bring a second wave of vertical AI consolidation beyond coding, medical scribing, and legal.

Elad predicts that after 2025 consolidated coding, medical scribing, and legal (Harvey and others) into a handful of dominant players each, the next tier of verticals will hit massive scale and consolidate the same way in 2026.

ai-adoption-and-verticals

The 'AI is overhyped' and 'AI isn't really working' narratives will keep recurring every year regardless of actual progress.

Elad argues new technologies generally take about a decade to fully propagate, so both the bubble narrative and the skeptical narrative will keep resurfacing in public discourse even as real economic value keeps compounding underneath, calling most of that discussion a waste of time.

ai-adoption-and-verticals

Consumer AI products have lagged despite clear technical readiness, mainly because too few people have both the product instinct and the current-technology context to build something genuinely new.

Sarah relays a well-known consumer founder's view that at most a few hundred people worldwide have the exceptional consumer product instinct combined with real understanding of what AI now makes possible; most attempts so far have just been AI-powered rebuilds of prior-generation products, but both hosts expect differentiated 'magical' consumer agent products to start emerging in 2026.

ai-adoption-and-verticals

Robotics in 2026 will follow a similar hype-then-real-progress curve to self-driving cars circa year 15-17.

Both hosts expect humanoid and semi-humanoid robots to reach small-scale consumer and industrial deployment in 2026, with visible early failures triggering an outsized negative reaction, followed by a longer runway before the technology is actually reliable, mirroring the multi-decade arc of autonomous vehicles.

robotics-and-autonomy

Capital-intensive, hardware-heavy robotics markets structurally favor incumbents over pure-play startups.

Elad argues Tesla (via Optimus), Google (via Waymo), and Chinese automakers barred from the US market are best positioned to win robotics because the field requires deep capital, manufacturing, and supply-chain expertise beyond just a good model; Sarah pushes back that some startups will still succeed, citing SpaceX as precedent for hardware startups building out an entire supply chain from scratch.

robotics-and-autonomy

2026 will bring a wave of AI company IPOs driven by retail demand to 'not miss the next Nvidia', more than by fundamentals.

Elad predicts strong retail appetite will push several AI labs and companies public and that a successful first mover will pull more issuers into the market. Sarah relays a hedge fund manager friend's admission that his fund will buy AI IPOs on game theory ('I can't miss it like I missed Nvidia') and benchmarking pressure, regardless of fundamental conviction.

capital-markets-and-ipos

The real risk to the AI capex cycle is financial-plumbing risk, not doubt that AI demand is real.

Sarah frames the 'AI bubble' fear as concern about who actually bears the credit risk in pay-on-delivery data center and chip contracts, and about concentration in Nvidia and a small number of other players, rather than genuine doubt that AI usage and value are real.

capital-markets-and-ipos

Ilya Sutskever's 'age of research' framing argues that a finite compute floor for testing ideas means algorithmic and architectural breakthroughs, not just raw scaling, will matter more going forward.

Elad paraphrases Sutskever's recent interview: scaling still matters, but there is a limit to how much compute exists to test new ideas at scale, so labs able to find more compute-efficient improvements (diffusion, SSMs, self-improvement, continual learning) are no longer purely in a resource-scale race.

research-frontiers

AI applied to hard sciences will produce a few overhyped 'one-off' breakthroughs in 2026 while the real long-term impact stays underestimated.

Elad predicts isolated wins in materials, math, or physics (a new material discovered, a conjecture proved) will get inflated into 'science is solved' narratives, following the same hype-then-underestimation pattern he sees repeat across AI subfields, before the underlying trend proves out more slowly and durably.

research-frontiers

GLP-1 drugs remain underrated in their second-order effects and are opening a broader path for engineered peptide and hormone therapies.

Elad argues GLP-1 adoption is inexorable given how effective it has been, and its success is normalizing biohacking-adjacent peptide, hormone, and longevity interventions (neuromodulation, stem cell injection) that were previously a fringe community, mirroring how bodybuilding-community supplements like creatine went mainstream.

beyond-ai

Defense tech and autonomous/drone-based warfare will accelerate faster in 2026, driven by both policy tailwinds and startup density.

Sarah predicts defense tech will accelerate due to Trump administration posture toward defense modernization plus a critical mass of startups now working in the space; Elad agrees directionally but notes budget still has to shift from prime contractors to new entrants before that shows up at scale.

beyond-ai

Media referenced

MIT report on AI ROI (widely-cited 2025 study claiming most AI pilots fail to show return) - paper - Elad dismisses this as part of the recurring yearly 'AI isn't working' narrative cycle, arguing it did not actually reflect the real adoption picture.
Ilya Sutskever interview on the 'age of research' - other - Elad paraphrases Sutskever's framing that scaling still matters but there is a finite compute floor for testing new research ideas, motivating a shift toward algorithmic and architectural breakthroughs.

Companies

OpenAI - ChatGPT cited for shipping massive user numbers and raising massive capital this year.
Google - Gemini cited as having 'roared back strong' in 2025; also the incumbent behind Waymo.
Nvidia - Central to the concentration-risk and capex-bubble discussion; running joke about employee compensation and market dependence.
Waymo - Cited as the incumbent (Google-backed) winner in self-driving, used as the analogy for how incumbents may win robotics too.
Tesla - Optimus and Tesla self-driving cited as likely incumbent winner in humanoid robotics.
Harvey - Cited as an example of legal AI vertical consolidation into a handful of players in 2025.
Abridge - Cited alongside Open Evidence as evidence of fast physician adoption of AI documentation and clinical decision support.
Open Evidence - Cited alongside Abridge as evidence of fast physician AI adoption.
DeepSeek - Its late-2024 release marked the start of the frontier of open intelligence shifting from the US to China.
xAI - Colossus data center cited as an example of power-constrained AI buildout initially powered by on-site gas turbines.
Reflection AI - Cited as one of the new 'neolabs' pursuing open-weight intelligence to help the US regain ground from Chinese open models.
SpaceX - Cited as precedent for a capital-intensive, hardware-heavy startup succeeding despite needing to build an entire supply chain, used to argue robotics startups can still win.

Techniques and frameworks

Age of research - Ilya Sutskever's framing (paraphrased by Elad) that progress is shifting from pure compute scaling toward algorithmic ideas, since compute for testing new research directions at scale is finite.
AlphaFold-style structural prediction as a precedent for AI research methodology - Elad analogizes protein design's shift from analytical design to iterative methods (phage display, mutagenic scans) and finally AI-solved structure prediction, as a possible precedent for how AI development itself could evolve toward more evolutionary, self-improving systems.
Agent harnesses and context engineering - Named as the missing scaffolding needed to get order-of-magnitude capability gains out of existing models when deploying agents in the enterprise.
Economically useful evals (e.g. Apex, GDPval) - Cited as the direction evals need to move in 2026, from benchmark scores toward measuring real knowledge-work economic value.

Summary

This is No Priors' year-end 2026 forecast episode: Sarah Guo and Elad Gil trade roughly a dozen predictions for the coming year, then the show hands off to a montage of quick predictions from past guests and friends of the show. The tone is casual and joke-heavy (a long running bit about microplastics versus silicon opens the episode), but the substance covers enterprise adoption, robotics, capital markets, research direction, and a couple of predictions that have nothing to do with AI at all.

On adoption, both hosts expect 2026 to bring a second wave of vertical consolidation beyond the coding, medical scribing, and legal verticals that already collapsed into a handful of winners in 2025. Sarah flags an underdiscussed pattern: historically conservative, slow-adopting professions like medicine, law, and accounting are now the fastest adopters of AI, citing physician uptake of tools like Abridge and Open Evidence. Elad predicts the "AI is overhyped" versus "AI isn't really working" debate (including a dismissal of a widely-cited MIT ROI report) will keep recurring every year regardless of underlying progress, since new technologies generally take about a decade to fully propagate.

On robotics and autonomy, the hosts compare 2026 to self-driving circa year 15-17: real but uneven progress, small-scale humanoid deployments that will not all work, and outsized reactions to early failures. Elad argues capital-intensive, hardware-heavy robotics favors incumbents like Tesla (Optimus), Google (Waymo), and Chinese automakers over startups, while Sarah counters with SpaceX as proof a hardware startup can build out a full supply chain from scratch. They also debate the definition of a "robot" at length (are dishwashers and elevators robots?), landing on intelligence and generalization as the dividing line.

On capital markets, Elad expects a wave of AI IPOs next year driven less by fundamentals than by retail demand to "not miss the next Nvidia," illustrated by Sarah's story of a hedge fund friend who plans to buy AI IPOs purely on game theory regardless of his actual view of the company. Sarah reframes the "AI bubble" fear as concern about financial plumbing, specifically who bears credit risk on pay-on-delivery data center and chip contracts, and concentration in Nvidia and a small number of other players, rather than genuine doubt that AI demand is real. On research, Elad paraphrases Ilya Sutskever's "age of research" framing: since compute available to test new ideas at scale is finite, algorithmic and architectural breakthroughs (diffusion, SSMs, continual learning, evolutionary self-improving systems) matter as much as raw scaling going forward. He also predicts a few overhyped, isolated AI-for-science wins (a new material, a proved conjecture) will be inflated into "science is solved" narratives in 2026.

The episode closes with each host giving one non-AI prediction (Elad on defense tech and drone-based warfare accelerating under policy tailwinds and startup density; Sarah on GLP-1's underrated second-order effects opening a path for broader peptide and hormone therapies), followed by a montage of rapid predictions from other guests and friends of the show covering reasoning models transforming every industry, proactive rather than reactive AI assistants, context and memory as the next product battleground, agent harnesses and economically useful evals, US open-model leadership, AI's growing politicization ahead of the 2026 midterms, and energy-efficient AI compute.

Notable Quotes

"The people who tended to be the slowest adopters of technology love AI. That's physicians, that's lawyers, that's certain accounting types. It's actually kind of fascinating." - Sarah Guo

"The reality of the technology is it takes like 10 years to propagate, and people are getting enormous value out of AI already and they're going to get way more out of it in the future." - Elad Gil

"They are concerned that demand isn't real for AI to support the capex cycle, that there is systemic risk from people passing the ball around in terms of who is actually responsible for the capex buildout." - Sarah Guo

"His view was like, you buy the IPO regardless of your fundamental view of the company. And I was like, wow, this is not the investing job I know how to do." - Elad Gil, recounting a hedge fund manager friend's reasoning on AI IPOs

"2026 prediction would be that this is going to be certainly the continued year number two of AI agents, but in particular AI agents in the enterprise in either deep vertical or domain specific areas." - Aaron