Companies Are Becoming a Series of Loops: AI Can Climb the Hill, but Choosing the Next Mountain Still Needs Humans
Companies are being reshaped into a cascade of interlocking loops; loops can push current businesses to local optima, but after the plateau, choosing the next mountain still relies on human intuition.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The AI era won't replicate winner-take-all dynamics
The AI era won't follow mobile internet's winner-take-all pattern: network effects once pushed search and social toward one or two players, but each layer of the AI stack now has about 20 relevant participants. Two years ago, everyone thought programming agents would end with a single winner; today Claude Code, Codex, Lovable, and Replit are all advancing simultaneously. Anish says the most cutting-edge people in labs believe what's actually happening is autocatalytic effects—using new technology to improve your own processes—rather than recursive self-improvement, so there won't be a runaway winner that swallows everyone. The lesson for founders: don't bet everything on one full-stack giant; each layer (models, agents, interfaces) still has a long-term position.
— Anish AcharyaCEOs want to build bigger companies, not fewer employees
Employees probably don't need to worry excessively about being replaced by AI, not because the technology isn't strong enough, but because CEOs are incentivized by growth, not shrinkage. Sundar doesn't want to run a more efficient $4 trillion company; he wants to build a $40 trillion one. A Google executive friend told Anish that the company hasn't laid off staff; instead, AI compressed a two-year roadmap into three months, and the challenge now is 'what to add to the roadmap.' When execution is no longer scarce, attention shifts from saving labor to finding new things to do for bigger goals.
— Anish AcharyaCompanies will become loops; the next step requires human decisions
The company form is shifting from a list of roles and departments to a cascade of loops: starting with automation of individual workflows, then functional loops, then crossing business units, and eventually loops running most of the company's operations. Loops continuously produce outputs that become inputs for the next layer, and some mechanism feeds signals back to the CEO—when to adjust business entities, when to change strategy. This mechanism helps companies climb to local maxima, but after reaching a plateau, human intuition is needed for out-of-distribution thinking to place the organization at the foot of the next mountain. Letting AI directly come up with an idea that can make a million dollars often fails because humans can set the direction; AI only sweeps along the established slope.
— Anish AcharyaUsers want to spend time, not save it
Silicon Valley's default is that AI's value is efficiency; Anish thinks this is a misjudgment: most people don't really want to save time, but want to spend it on things that bring connection, love, progress, and fun. Historically, the biggest products have been entertainment and social; the consumer AI opportunity is therefore not 'saving a few more minutes' but creating loops that make people feel more connected, more loved, or experience progress or fun. He criticizes that technology over the past 40 years has extended intellect but hardly touched the soul; fitting AI into these desires is a product design challenge, not a model capability challenge.
— Anish AcharyaAI is decoupling skills from desires
Anish believes AI is not just a productivity tool but more like an emotional and spiritual interface—it can separate 'skills' from 'desires.' People who want to make music are no longer blocked by 'not knowing how to play piano'; those who want to write programs don't need to learn to a professional level first; personal identity is thus amplified. He recalls that high-risk environments in the 1950s-60s greatly stimulated human potential, and believes the current world is in a low-risk state; AI has the chance to lift GDP growth from 2% to 10-15%—provided it increases not just the speed of individual steps but the scale of what groups dare to imagine.
— Anish AcharyaIdeas too small now fail to raise funding
Investment criteria have reversed within three years. In the past, if a project was too ambitious, funds would often say a $100 million seed round was excessive; today at a16z, the reverse standard applies—if an idea can only be that small even with AI's help, they simply don't invest. 'Unbounded ambition' has become the core value for screening companies; Marc Andreessen's phrase: we either go to the moon or leave a moon-sized crater.
— Anish AcharyaShip something small weekly to build product intuition
The most direct advice for product people is an old saying: build more. Pick a project you tell no one about, the smaller the better, treat it as a 'chassis,' run each new model through it, ship the result, and discuss it with people—this is the path to building AI intuition. As long as you ship one small thing weekly, madness is unsustainable. Anish himself used Codex to make a Mother's Day slideshow for his wife: pulling material from texts and photo albums, auto-scoring music, and finally generating a 20-page retrospective of their relationship, even digging up old texts from their first date. The thing is small, but such works let you repeatedly taste AI's 'joy moments'—once tasted, your perspective never goes back.
— AnishIn their own words · checked verbatim
The loop will help you climb to the local maxima, but then it plateaus. You need human intuition. You need somebody to actually help you land at the base of the next hill.
Anish Acharya15:12
We believe that people want to be more productive, but they don't. I think more people want to spend time than save time.
Anish Acharya32:24
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.
Anish Acharya37:29
It can be totally unimportant. Um but use it as a chassis to use all the new models, ship things, talk about them, build your own intuition.
Anish1:09:48
Figures
| Number of relevant players per layer of the AI stack | about 20 | 4:05 |
| CEO target company size comparison | $4 trillion → $40 trillion | 10:08 |
| Roadmap execution time compression | 2-year roadmap completed in 3 months | 10:08 |
| Share of administrative costs in healthcare spending | 45% | 40:31 |
| Potential GDP growth driven by AI | 2% → 10-15% | 37:29 |
Glossary
- loop
- A workflow that continuously iterates using its own output as feedback; Anish uses it to describe the minimal operating unit of future companies.
- autocatalytic effects
- A mechanism where using new technology to improve one's own processes creates acceleration rather than runaway amplification.
- recursive self-improvement
- AI repeatedly improving itself leading to exponential runaway; Anish argues this is not the dominant mechanism at present.
How to listen
Founders and product leaders using AI to reshape their products or reorganize company processes, as well as investors concerned with AI's impact on employment and consumer opportunities.
If short on time, you can skip the final book list, life mottos, and DJ anecdotes without missing the AI analysis.