Why companies are becoming a series of loops | Anish Acharya (a16z)
with Anish Acharya
6 Sep 20265 min read1h 20m
TL;DR
Anish Acharya argues that the 'permanent underclass' fear around AI is Silicon Valley's collective dark fantasy — empirically unfounded, since job postings are up and AI progress is autocatalytic, not recursively self-improving. He introduces the 'loops' framework: every business function (engineering, growth, legal, sales) is becoming an agent loop that climbs to a local maxima, then requires a human to reset direction toward the next hill. The biggest consumer opportunity isn't productivity but the basics of human need — feeling connected, loved, and having fun.
Key Moments
Anish Acharya
“We believe that people want to be more productive, but they don't. I think more people want to spend time than save time. So, I think that the opportunity for this technology is the basics of consumer need. How do we feel more connected, more loved? How do we make progress? How do we have fun?”
Anish describes what he sees as the real consumer opportunity for AI — not productivity tools but fulfilling fundamental human needs.
“It's not actually RSI that's occurring, which could lead to some sort of runaway winner cuz they were an epsilon ahead of the others. It's autocatalytic effects, which just means you're using the new technology to improve your process, but it's not truly recursive.”
Anish distinguishes between recursive self-improvement (RSI) and autocatalytic effects when explaining why AI won't produce a runaway winner.
“How many problems are truly intelligence bound? Like if you had a, you know, a data center of PhDs working at FedEx or Domino's Pizza, are they going to be like exponentially dominating supply chain and pizzas? Like I don't think so.”
Anish challenges the assumption that more AI intelligence automatically translates to outsized business outcomes across all industries.
“The loop will help you climb to the local maxima, but then it plateaus. And you need some sort of out of distribution thinking, you need human intuition, you need somebody to actually help you land at the base of the next hill.”
Anish explains the core limitation of agent loops in business functions — they optimize well but can't break out of local maxima without human direction.
“I feel like every PM at every company feels like they're a true zero-to-one thinker, but they're held back by the kind of, you know, the heavy hand of management and executives and founders and engineering capacity. And in a world where every story gets told, every product story gets told, every feature gets tried, I think a lot of PMs are going to realize they're actually not that good at zero-to-one.”
Anish reacts to a PM friend's idea of saying yes to every feature idea and simulating them all, suggesting AI will expose which PMs are actually strong zero-to-one thinkers.
Anish Acharya is a General Partner at Andreessen Horowitz (a16z) where he focuses on consumer investing. He is a long-time product builder and founder, having founded Social Deck (sold to Google) and Snowball (sold to Credit Karma), where he later became VP of Product and GM of the consumer product and credit card business. He has been at a16z for over 7 years and brings a rare operator-investor perspective to AI and consumer product strategy.
Takeaways
1
Kavak's 'agent calls human' model is the template for human-AI loops Kavak runs one agent per customer for used-car sales in Mexico; when the agent gets stuck, it calls a human who coaches it through. The agent captures the interaction traces and learns, so it shouldn't need to call a human for that scenario again. This is a concrete, replicable pattern: treat every agent failure as either a knowledge gap or a data gap, then close it.
2
Every function is becoming a loop; humans reset direction at plateaus Anish's 'loops' framework: agent loops (bug fix → repro → fix → review → ship) will expand from engineering to growth, legal, sales, and support. These loops climb efficiently to a local maxima then stall — a human must provide out-of-distribution thinking to point toward the next hill. The practical implication is that every team lead should map their function as a loop and identify where human judgment is the rate-limiting input.
3
Frontier models only rational for unbounded-upside jobs Anish applies Pareto efficiency to model selection: for drug discovery or sales, one extra IQ point could unlock trillion-dollar outcomes, justifying frontier model costs. For closing the books or routine legal work, you can only be correct — there's no 100x upside — so a cheaper, fine-tuned open-weight model is the rational choice. Companies that default to frontier models for every function are leaving significant cost efficiency on the table.
4
Not all problems are intelligence-bound — know which ones are Anish's key provocation: a data center of PhDs at FedEx or Domino's wouldn't exponentially dominate pizza or supply chain, because those problems aren't intelligence-bound. The strategic question for any company is whether their core value creation is intelligence-bound or constrained by something else entirely. Founders and executives should explicitly audit this before over-investing in frontier AI capabilities.
5
AI progress is autocatalytic, not recursively self-improving The most sophisticated researchers at frontier labs distinguish between autocatalytic effects (using AI to improve your process) and true recursive self-improvement (RSI). Anish argues RSI is not actually occurring, which undercuts the 'runaway winner' scenario and the permanent underclass fear. This is a technically important distinction that changes how urgently you need to react to every new model release.
6
AI will reveal which PMs are actually zero-to-one thinkers When every feature can be prototyped and user-tested cheaply at scale, the excuse that 'management blocked my ideas' disappears — the best idea simply wins in testing. Anish predicts many PMs will discover they're not as strong at zero-to-one as they believed, and will find more fulfillment executing someone else's better idea. This is a rare honest take on a professional identity shift that most PM discourse avoids.
7
The real consumer AI opportunity is connection and fun, not productivity Anish argues that consumer AI products are collectively chasing productivity when most people actually want to spend time, feel connected, feel loved, and have fun. This isn't a model or capability gap — it's a product design challenge. Founders building consumer AI should reframe their value proposition around these primary human motivations rather than time-saving.