Lenny's PodcastWhy companies are becoming a series of loops | Anish Acharya (a16z)
At a glance
WHAT IT’S REALLY ABOUT
AI will turn companies into automated loops guided by humans
- Anish Acharya argues that fears of AI creating a “permanent underclass” are overblown because the AI ecosystem is unusually competitive and decentralized across models and products.
- He predicts AI’s biggest near-term impact will come from reorganizing work into automated “loops” that handle end-to-end workflows across functions, while humans remain critical for strategy, judgment, and exceptions.
- He expects a bifurcation in AI usage inside companies: frontier models for high-upside or ambiguous work and cheaper open-weight or optimized models for bounded, verifiable tasks.
- In consumer, Acharya believes the major opportunity is designing AI products that improve happiness and connection (“loop make me happier”) rather than narrowly optimizing productivity.
- On moats and distribution, he claims classic defensibility still works, but most moats are discovered through shipping high-craft products that earn organic word-of-mouth.
IDEAS WORTH REMEMBERING
5 ideasThe “permanent underclass” fear is mostly a Silicon Valley dark fantasy.
Acharya argues the last tech wave centralized power via network effects, but today’s AI stack has many viable players across labs, open-weight models, and application layers (e.g., multiple coding agents succeeding simultaneously). The data and market structure suggest broad access and many winners rather than a single runaway incumbent.
AI progress can be fast while economic change remains slow.
He’s skeptical of a sudden, uncontrollable intelligence “jump,” noting that diffusion into the real economy is slow and many business problems aren’t primarily limited by raw intelligence. Even big model progress doesn’t instantly translate into overnight societal or company-level transformation.
Winning companies will be built as layered, cascading “loops,” not one-off AI features.
He describes the evolution from prompts → agents (models in loops with tools/memory) → organizational “loops” that automate repeatable workflows end-to-end (engineering, growth, support, sales, legal). The aim is cascading loops: per person, per function, up to business-unit or company-level loops that surface strategic changes.
Agents hill-climb to local maxima; humans reposition to the next hill.
Loops optimize toward measurable improvements (e.g., rapid experimentation), but eventually hit a plateau; moving to a new “hill” requires human taste, intuition, and out-of-distribution judgment. Humans remain essential for strategy, exception handling, and setting direction—even in AI-native orgs.
Organizations will run a mixed “model portfolio,” not one model for everything.
He expects a split where some tasks warrant expensive frontier intelligence (high upside, ambiguous, or where small gains compound massively—research, engineering, sales/support) while other bounded-upside functions can be served by cheaper, optimized, often open-weight models. “Best model” isn’t universal; comparative advantages across model families matter in practice.
WORDS WORTH SAVING
5 quotesIts a funny dark fantasy that we seem to have as, you know, Silicon Valley collectively. Like, things have never been better really by almost every measure... And yet there's this sort of discussion of permanent underclass... and yet we can't seem to let go of this fantasy.
— Anish Acharya
Were gonna 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. With that said, I think humans are a critical ingredient.
— Anish Acharya
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 Acharya
We built this, like, technology that extends our intellect, but nothing to extend our soul.
— Anish Acharya
I don't think it's a model or a capability challenge, it's just a product design challenge.
— Anish Acharya
High quality AI-generated summary created from speaker-labeled transcript.
