Why AI will dwarf every tech revolution before it: robots, manufacturing, AR glasses from CES 2026
Summary
AI is compressing product cycles and value creation enough to dwarf the PC, internet, cloud, and mobile eras. Bob Sternfels calls the pace “literally warp speed,” while Hemant Taneja describes “peak ambiguity”: technology capabilities and geopolitical conditions are changing simultaneously, making organizational speed more important than fixed plans.
Anthropic is the episode’s clearest specimen of this compression. Taneja says the business was doing roughly $880 million when General Catalyst invested, after 10x growth, and later announced another 10x-or-more year. He framed the $60 billion valuation against an approximately $8 billion-$10 billion run-rate business as “the cheapest deal that got done last year” on financial metrics. Venture investors must now consider whether trillion-dollar companies are plausible.
Enterprise adoption and IT spending can drive model-company growth, but nontechnology companies are still struggling to realize value at scale. Calacanis dramatizes the CFO-CIO conflict: one sees spending without ROI, while the other warns that delay invites disruption. Sternfels sees a path to make them allies; Taneja points to data infrastructure, adapted models, and redesigned human-agent workflows rather than isolated pilots.
General Catalyst is buying some declining incumbents as transformation and distribution infrastructure for AI startups. Taneja describes the firm as “venture capital for America,” meeting founders at different stages with flexible capital, policy capabilities, and market access. Its Ohio health-system acquisition provides a site to deploy AI and demonstrate a healthcare playbook; declining call-center assets can similarly provide access to customers. Sternfels calls this a new asset class: transformation rather than conventional private-equity optimization.
AI is splitting headcount growth, output, and organizational layers rather than shrinking every role uniformly. At McKinsey, Sternfels says client-facing staff will grow 25% next year while the non-client-facing half of the firm shrinks 25% and increases output 10%. With 40,000 humans and 25,000 personalized agents—and expected parity by year-end—the firm is simultaneously adding and shrinking.
The entry-level labor market is losing its old training bargain, making initiative and human judgment more valuable. Calacanis says graduates sending 100-200 résumés may receive no offers because training someone can take longer than building an agent; he recommends direct outreach and useful spec work demonstrating “chutzpah,” drive, and skill. The panel places human value in leadership and goal-setting, judgment and parameter-setting, creativity, curiosity, and resilience.
Physical AI will be constrained as much by manufacturing economics as by model intelligence. Sternfels expects a major autonomous-vehicle transformation over the next 12-24 months and cites 50,000 unfilled U.S. manufacturing jobs at one contract manufacturer, while Taneja warns robotics will diffuse more slowly than LLMs because there is no equivalent hardware API. Calacanis nevertheless predicts Tesla’s Optimus 3 will eclipse its cars, reach billion-unit scale, and become “the most transformative technology product ever made.”
Today’s awkward interfaces may be transition products toward continuous health intelligence and less intrusive computing. Google Glass had an early AR direction but insufficient utility; Taneja sees wearables, blood testing, and tiny-volume diagnostics as possible precursors to customized medicine. Calacanis compares LLM hallucinations to a Discman’s skipping. Sternfels’s broader behavioral bet is renewed in-person connection rather than trying to be fulfilled online while lonely offline.
Deep dive
1. AI has compressed the clock on product cycles and value creation
Calacanis’s opening claim was categorical: everything from PCs and the internet to mobile and cloud “is going to be dwarfed” by AI’s social impact. Sternfels’s CEO conversations support the urgency—technology is now central to every industry, and the recurring question is how to make organizations move faster.
Taneja’s framing is “peak ambiguity”: geopolitical alliances, national drives for strategic autonomy, and the underlying implementation technology are all changing together. Builders must create enduring value with tools that may themselves obsolete today’s product architecture.
The venture comparison captures the new clock speed. Stripe, an investment from 2010, took roughly 12-13 years to become a $100 billion company; Anthropic moved from a $60 billion valuation last year toward “a couple hundred billion” far faster, alongside what Taneja described as real business growth.
Anthropic’s application layer matters as much as its models: Taneja sees Claude transforming enterprise engineering. His numbers—about $880 million after one 10x year, followed by another announced 10x-or-more year—led him to ask whether venture capital should now aim beyond decacorns toward trillion-dollar companies.
2. Enterprise AI must escape pilot purgatory
Sternfels attributes model-provider growth to large enterprises adopting AI and raising technology spending, but he preserves the catch: realizing value at scale inside nontechnology companies “is proving harder than people think.”
Calacanis dramatizes the CEO’s dilemma as a CFO-CIO standoff. The CFO sees spending without ROI and asks to pause; the CIO says delay invites disruption. Sternfels’s response is that there is a path to make the two functions allies.
Taneja says successful transformation requires three connected layers: enterprise-ready data infrastructure, models adapted to the business, and a workforce model for humans and agents. Department-level applications—from coding and call centers to sales, marketing, and healthcare—only matter once people can absorb them into daily work. He cites Transcarent’s AI-enabled healthcare access, including routing for surgery, cancer therapy, or mental health, as an example of a broader workflow transformation.
3. Buying incumbents can become an AI distribution strategy
Taneja says General Catalyst remains “venture capital for America,” meeting founders where they are across a company’s life cycle. Its support includes flexible capital, policy capabilities, market access, and global relationships; the Ohio health-system acquisition is an extension of that strategy, not a departure from the seed business.
He explains the Akron, Ohio health-system acquisition as market access, not conventional private equity. After converting the nonprofit with the attorney general, the firm intends to deploy founders on the ground, improve care and resilience, then carry the playbook to hundreds of other systems. Taneja stresses that the hospital must continue operating and serving its community.
Call centers offer the same mechanism: their existing labor model may be a declining asset, but their customer relationships remain valuable. Buying one lets an early-stage founder work with those customers, accelerate AI adoption, and scale faster.
Calacanis’s metaphor was “buy the castle, open the drawbridge”: venture capitalists once financed barbarians attacking incumbents, but can now purchase the incumbent and install startups inside it. Sternfels agreed this resembles a new asset class—transformation rather than optimization—with enterprises facing a choice to “transform or die.”
4. AI is bifurcating organizations rather than shrinking them uniformly
Sternfels summarized McKinsey’s plan as “25 squared”: grow client-facing staff 25% next year while reducing the non-client-facing half of the firm by 25%, with that group producing 10% more output. The firm can now grow capability and total headcount without every part of the organization growing together.
McKinsey saved 1.5 million hours in search and synthesis last year, but Sternfels says those hours are being “dividended” into harder client problems. Agents also generated roughly 2.5 million charts in six months—an output he joked made him want to eliminate charts altogether.
As of the prior week, McKinsey had 40,000 humans and 25,000 personalized agents, with parity expected by year-end. Sternfels identifies structured problem-solving, search and synthesis, and communication as areas where the technology is working particularly well, while life-and-death healthcare decisions still call for humans because the technology is not yet reliable enough.
Taneja sees the same shift in startups: when “code self-writes,” innovation becomes less about writing code quickly and more about embedding technology into systems. Founders need constant iteration, trusted customer relationships, and “radical collaboration,” because neither side knows exactly where the capabilities will lead.
5. Human advantage moves toward aspiration, judgment, and resilience
Asked what models cannot do, Calacanis offered leadership and goal-setting. Taneja named judgment: humans must set parameters and evaluations based on firm values and societal norms, as well as provide true creativity that is not merely the next likely inference. Sternfels emphasized asking better questions, imagination, curiosity, and the ability to shape the world toward a compelling vision.
Taneja translates that into education: a world with abundant problem-solving technology rewards asking the right questions, curiosity, and imagination. The relevant pedagogy looks more like Socratic dialogue and less like every seventh-grader factoring polynomials at the same scheduled hour.
Calacanis’s labor-market warning was harsher: graduates may send 100-200 résumés without an offer because employers conclude that training a junior worker takes longer than building an agent. His alternative is to email the CEO with useful spec work—such as a redesigned landing page—and demonstrate skill before asking for a formal pathway.
Taneja says this may also widen the talent funnel: where someone went to school could matter less than intrinsic qualities and evidence such as a candidate’s GitHub profile. Sternfels added resilience as the missing institutional capability: “You’re going to get knocked down. The question is, do you get back up?”
6. Lifelong learning must replace the four-year credential bargain
Taneja calls the sequence of learning for 22 years and working for 40 “a broken idea.” His alternative is lifelong college: an enduring relationship through which people continually skill and reskill as technology and job definitions change.
Sternfels supplied the economic reason. An employer’s expected return period on an employee’s skills has fallen from about seven years to roughly 3.6 years over three decades, making the ability to learn repeatedly more valuable than mastery of one static subject.
Taneja likened workers to moving from orchestra members toward conductors, and Calacanis extended the metaphor to people directing their own ensembles of agents. At a Singapore dinner, all 12 founders said an LLM wrote recent job descriptions; about half had built agents to sort and stack-rank applicants.
Sternfels says every department at McKinsey now needs AI teammates, but whether an agent is a copilot or autonomous pilot depends on reliability, complexity, and severity. In healthcare, he says humans should make life-and-death decisions today. Taneja’s warning is that eliminating the bottom four rungs of the ladder may save money now while removing the pathway to the organization’s future CEO.
7. Robotics leadership depends on rebuilding manufacturing
Calacanis dubbed 2026 “self-driving CES,” citing Waymo’s lead and progress from Tesla, Zoox, and Nuro/Lucid, while predicting consumer humanoid robotics becomes the 2027 theme. Taneja widened the lens: BYD and other Chinese companies are penetrating Europe and the Middle East with rich functionality at low cost.
The U.S. may possess strong self-driving innovation without the manufacturing economics needed for mass adoption. Taneja argues AI-enabled design and production—including work at Re:Build Manufacturing—must close that gap, because self-driving that cannot reach the right price point remains a limited product.
Sternfels expects “a Western stack” and “a Chinese stack” to compete over the next 12-24 months. Robotics also addresses labor scarcity: one contract manufacturer has 50,000 unfilled U.S. jobs, while Korea has roughly one robot per ten workers; Germany and China are tied behind it, and the U.S. is a distant third.
Taneja’s pushback is that robotics will be “slower than people think”: good models cannot spread through physical infrastructure as easily as ChatGPT spread through the cloud. Calacanis took the opposite extreme after seeing Optimus 3, predicting one robot per human, a billion units, and a future in which nobody remembers Tesla primarily for cars.
8. Awkward gadgets reveal where interfaces are heading
Sternfels recalled a project “we” did for AT&T in the mid-1980s that concluded cellular phones would not take off—a useful admission of forecasting failure. Google Glass presented the inverse problem: directionally early, but newer glasses still have better form factors without enough utility, according to Taneja.
Calacanis presented the Theranos one-drop device as a compelling product promise while hedging his characterization of the company. Taneja thinks accurate, tiny-volume diagnostics are “very likely” within ten years if nanodevice manufacturing advances, enabling continuous and preemptive healthcare rather than occasional snapshots.
Taneja sees Oura, Whoop, Eight Sleep, wearables, and blood testing as transition technologies toward customized medicine. Calacanis offered the software analogue: today’s LLM hallucinations may eventually look like a Discman’s skipping—an obvious defect of an intermediate product, not the defining property of the mature category.
The pager represented the always-on work trajectory that Calacanis linked to doomscrolling; now some consumers are unbundling smartphones into digital cameras and flip phones. Sternfels’s hoped-for behavioral reversal is renewed human connection: more in-person engagement rather than trying to be fulfilled online while lonely offline.