Add to Google Preferred Sources
a16z general partner Anish Acharya, speaking on Lenny’s Podcast, argues Silicon Valley’s fear of a “permanent underclass” is a dark fantasy, and that the real story of AI is the unbundling of skill from desire. He lays out a framework where companies become cascading loops of AI agents that plateau at local maxima, requiring humans to provide out-of-distribution thinking to find the next hill. On the consumer side, Acharya contends the industry is stuck in an “iPhone 2010” moment, having focused too heavily on productivity tools instead of applying AI to fundamental human needs like connection and happiness. His most striking prediction is macroeconomic: by stripping administrative waste from sectors like healthcare—where 45% of costs are administrative—and reorienting society toward ambition, AI could drive US GDP growth from today’s 2% to 10%, 15%, or even 20%. For founders, he offers a contrarian playbook: stop asking how to make a free product, and instead ask what the $1,000-per-month “software Birkin bag” version would look like.
Key Elements
Silicon Valley has spent the last two years nursing a collective anxiety attack about AI replacing workers, creating a permanent underclass, and spiraling into a runaway intelligence explosion. Anish Acharya, a general partner at a16z, thinks this is nonsense. In a wide-ranging conversation on Lenny’s Podcast, the former founder who sold Social Deck to Google and Snowball to Credit Karma dismantled the displacement narrative and replaced it with a far more provocative thesis: the biggest constraint on AI’s economic impact isn’t model capability or compute—it’s our own lack of ambition.
“It’s a funny dark fantasy that we seem to have as Silicon Valley collectively. Like things have never been better by almost every measure,” Acharya said, pointing to the structural differences between today’s AI stack and the mobile era that preceded it. Where Web 2.0 produced winner-take-all network effects, the AI landscape has dozens of relevant players at every layer—labs, open-weight models, and coding agents like Claude Code, Codex, Lovable, Replit, and Wabi all thriving simultaneously. Radiologists, he noted, have been “cooked” for 20 years by predictions of AI replacement, yet job postings are at all-time highs. The same holds for programmers.
His deeper point is about how the technology actually diffuses. Acharya rejects the recursive self-improvement (RSI) thesis that animates fast-takeoff scenarios. The labs’ most sophisticated thinkers, he argues, describe the current dynamic not as runaway recursion but as autocatalytic effects—using new technology to improve process without true exponential self-amplification. And even if intelligence were the bottleneck, how many real-world problems are actually intelligence-bound? A data center of PhDs at FedEx or Domino’s would not exponentially dominate supply chains or pizza delivery. Economic diffusion, he reminds listeners, is inherently slow. He grew up in a small town, and people’s lives there haven’t changed much despite all the technological progress.
Companies as Cascading Loops
The episode’s most substantive framework concerns how companies will reorganize around AI. Acharya traces an evolution from prompts to agents (models in a loop with tools and memory) to loops (sets of agents performing tasks), and finally to cascading loops that span entire business functions.
Coding is the proving ground because it has the best models, the most technically capable customers, and established loop patterns. A bug-fix loop takes a report, generates a repro, creates a fix, reviews it, and ships it—with human confirmation only for high-risk changes. The question Acharya poses is what the business loops look like: a general manager overseeing loops in coding, marketing, sales, support, and legal, where the output of all those loops is itself a loop that can be optimized.
“We’re going to see this sort of cascading set of everything from a loop per person, loop per job function, loop across entire business units to, you know, loops that can run large parts of the company,” he predicted.
The critical insight is that loops plateau. They climb to local maxima efficiently, but reaching the next hill requires human intuition and out-of-distribution thinking. Acharya illustrates this with a simple test: asking an agent to “make me a million dollars, make no mistakes” doesn’t work because someone must set the direction. The most powerful example comes from Kavak, the Mexican used-car company, where the CEO runs an agent per customer. When an agent gets stuck, it calls a human who coaches it through—and the agent captures the traces and learns, so the next time it doesn’t need to call. This creates a flywheel where human intervention systematically shrinks over time.
The Model Sommelier’s Guide to Late 2026
Acharya has earned a reputation as a “model sommelier” through a disciplined habit: shipping something with every new model that releases. His key claim is that models are not fungible commodities—they have distinct “shapes” of intelligence that suit different tasks.
| Model | Character | Best For | Notable Limitations |
|---|---|---|---|
| Qwen 38B | Creative, long-horizon, great storyteller | Generating 4-5 hour creative projects (e.g., documentaries with video via MiniMax and audio via ElevenLabs) | Less precise for structured product work |
| GLM 53 | Neurotic, precise, PhD-like | Product work requiring rigor | No vision component |
| Grok (model) | Ambitious, unhinged | Personal agent tasks, credential caching | Emerging from a non-traditional lab |
| Frontier models (Mythos, etc.) | Highest raw intelligence | Unbounded-upside problems (drug discovery) | Irrationally priced for most tasks |
His practical advice for building intuition is to maintain a “chassis”—a project, ideally not important, that you continuously improve with each new model. He recommends shipping something weekly, however small. His example: using Codex to build a Mother’s Day slide deck for his wife that pulled from text messages and photo galleries, set to music. The point is that learning comes through doing, and shipping bad ideas is how you discover good ones.
Consumer AI Is Stuck in “iPhone 2010”
Acharya’s most contrarian position is that the productivity focus of AI is misplaced for consumers. People don’t actually want to save time; they want to spend it. The biggest products in history are entertainment and social, and AI’s consumer opportunity lies in addressing fundamental needs: feeling connected, loved, making progress, having fun.
“We built this technology that extends our intellect, but nothing to extend our soul,” he said.
He diagnoses the current consumer AI experience as a product design failure, not a capability gap. The “X AI user” is fearful, opinionated about model versions, and deeply pilled; the “Instagram AI user” sees it as a slightly better Google search. The gap between them is interface and design, not model quality. Three structural problems have held back consumer AI, which he analogizes to being in “iPhone 2010” (pre-Airbnb, pre-Uber, pre-WhatsApp):
- Cost: Models have been too expensive for free-to-use products; open-weight models are changing this.
- Interface: Chat works for the highest-agency people (Elon, Sam) but the average consumer’s ideal interface is TikTok—something between chat and short-form video is needed.
- Focus: The technology has been oriented toward productivity rather than connection and entertainment.
The three big consumer areas he’s watching are coding agents (which are becoming a general interface for interacting with the world, not just making software), personal agents (Grok Bot, ChatGPT Work, and startup Instinct), and entertainment/companionship products. Notably, he corrects a common assumption: the majority of companion product users are women in their 40s and 50s, not young men.
The New Economics of Startups
On durability, Acharya cites Decagon’s Jesse: “Moats are most often discovered, not designed.” Cursor is the exemplar—criticized for lacking a moat, it captured reasoning traces, trained its own models (Composer 1 and 2), and compounded advantages over time. The classic moats from Hamilton Helmer’s 7 Powers still apply; what’s changed is that software difficulty is no longer a barrier, so founders must build toward multiplayer, network-effect products.
On distribution, he challenges the assumption that incumbents have an insurmountable advantage. Gemini, despite Google’s massive cross-selling, is not winning. The current moment feels like “Christmas 2009” when everyone had a new iPhone and wanted to download apps—the window is open for startups willing to build in directions incumbents are uncomfortable with (companionship being the prime example).
His most provocative advice for founders is to invert the pricing question. Instead of asking how to make a free product, ask what the “software Birkin bag” version of your product would be—a $1,000 or $10,000-per-month product that justifies its price through ambition and craft. Price, he argues, is a measure of product-market fit. “Nobody has a growth problem these days, they have a product problem,” he said.
Ambition as the New Differentiator
A recurring theme is that AI unbundles skill from desire, enabling anyone to pursue ambitions previously gated by technical ability. “It kind of unbundles skill from desire. For example, if you want to make music, you can make music now. You don’t have to know how to play the piano,” Acharya explained.
This has profound implications for how companies should think about hiring, product scope, and even societal fulfillment. Acharya notes that the old VC wisdom—avoiding companies whose ambitions were too large—has inverted. Three years ago, a $100 million seed round was nonsensical because no problem justified it; today, an idea that’s too small is disqualifying. He quotes Mark Andreessen’s framing from when a16z raised its first fund: “We were going to the moon or we were going to leave a moon-size crater in the ground. There was no other option.”
He also observes that the 1950s and 1960s were eras of extraordinary ambition because the stakes were high (post-war mobilization). “When the stakes are high, we are awesome. When the stakes are low, we are at our absolute worst,” he said. When stakes are low, society atrophies into unproductive side projects. AI, by making important things cheap, raises the stakes and restores ambition.
He specifically calls out healthcare and education as the two areas where AI can have the most transformative societal impact. “If you look at health care, 45% is administrative. So, if you take a lot of that administrative burden out, you can actually see deflationary health care costs,” he said. Education now faces its strongest competition in 200 years, unbundling learning from institutions and status from credentials.
The 20% GDP Question
Acharya’s most striking prediction is macroeconomic. He argues that AI can drive US GDP growth from the current 2% morass to 10%, 15%, or 20% by dramatically driving productivity and ambition. It’s a claim that puts him in the same ballpark as Elon Musk’s recent G20 prediction of 20% to 30% GDP growth from AI and robotics, though the two arrive at the number from different angles. Musk emphasizes physical robots—10 billion humanoids within a decade—while Acharya’s focus is on administrative overhead, consumer products, and the psychological unlock of ambition.
The tension Acharya acknowledges is whether the slow rate of economic diffusion will keep pace with the rapid rate of technological progress. The “software Birkin bag” economics of expensive consumer products may scale beyond the early-adopter elite, or it may not. But his central thesis is clear: AI’s ultimate value is not efficiency but amplification of human agency. The “permanent underclass” fear inverts the actual dynamic—the technology is democratizing capability, not concentrating it. The loops framework shows that while AI will handle the climbing of local maxima across every business function, humans remain essential for the discontinuous jumps to new hills. And the startup landscape rewards ambition precisely because the cost of execution has collapsed—what remains scarce is vision, taste, and the willingness to ship. The unresolved question is whether a society that has spent two decades in low-stakes comfort can rediscover the high-stakes ambition of its postwar predecessors, and whether the technology will be enough to get it there.
Full content available at:Why jobs are becoming a series of loops | Anish Acharya (a16z)
- Anthropic Reportedly Delays IPO Timeline: Prospectus Pushed to Late September, Roadshow to Kick Off Mid-October
- Musk Tells G20 Summit: AI to Boost Global Economy by 20%-30%; Humanoid Robots to Reach 1 Billion Units Within a Decade
- Rene Haas: The Data Center Buildout, Not Chips, Is AI’s Next Bottleneck
- US and China Race to Schedule First AI Safety Dialogue Before Trump-Xi Summit
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
