AIANALYSISPODCAST
Anish Acharya says companies are shifting from functional org charts to small teams that own a metric end to end. The hard part is climbing between metrics.
The traditional functional org chart assigns a person to a team and a team to a function. The "loops" framework Anish Acharya, a general partner at Andreessen Horowitz, laid out on Lenny's Podcast this week reorders the unit: a small, autonomous team that owns a metric end-to-end, with the agentic work and the human judgment both inside the loop.
Acharya describes three waves: prompts, then agents, then loops. A prompt is a one-off ask. An agent can take a task and run. A loop runs on a metric: it watches the input, takes the action, measures the result, and adjusts. "The loop will help you climb to the local maxima," Acharya says, "but then it plateaus. You need human intuition. You need somebody to actually help you land at the base of the next hill."
That hill-to-hill handoff is the part most "AI will replace teams" takes skip. Acharya's favorite example is a coding loop: a bug report triggers a repro, the agent writes a fix, runs the tests, ships the low-risk path, queues the high-risk one for a human reviewer, and emails the customer. "That happens in five minutes," he says. The same shape, he argues, fits a growth loop: every variant generated, measured, converged on a Statsig P-value, and shipped with a long-term holdout, with humans stepping in only to escape a local maximum.
The reorganization argument borrows from electrification. "It took 40 years for us to get from the inception of electricity to reorganizing factories," Acharya says, "and that means like burning the buildings down and starting from scratch versus taking what was previously coal and simply swapping it with electricity." The point isn't the number. It's the two paths. Swapping coal for electricity inside a 1920s factory is cheap and leaves the building alone. Reorganizing the factory floor around electric drives is what actually pays off, and it took decades because the building had to be rebuilt.
A counterweight to that slow clock sits in the same conversation. When Acharya looks at the AI stack, he sees roughly 20 serious players per layer, including Claude Code, Codex, Lovable, Replit, and Wabi, and reads that as a sign AI is not consolidating the way mobile did. More competition per layer should mean faster diffusion, which should compress the 40-year clock. He doesn't say by how much, but the implication sits there.
Anecdotally, the compression is already visible. A Google executive told him the company didn't use AI to lay people off; it used AI to rip through its roadmap. "Two years of roadmap happens in three months," the executive said, according to Acharya. The single-source nature of that quote, an anecdote relayed by a friend rather than a Google filing, is worth flagging before anyone builds a thesis on it.
So is the framework itself. "Loops" is friendly language for a harder set of moves. It can mean a small team now owns a metric it didn't own before, which is real power. It can also mean the management layer between the team and the C-suite gets thinner, accountability for failures diffuses into the loop, and the humans "stepping in to escape local maxima" turn out to be a handful of senior people who already had the context. Acharya himself notes that the next stage of loops, sometimes called "graphs" because loops call other loops, is already being memed on X, including by people who think the framing is doing more rhetorical work than the model has earned.
Kavak, the used-car marketplace, gives a concrete test. It runs what Acharya calls a Jedi Academy for mechanics, a six-week course that ends with employees shipping a cutting-edge in-production agent. The test is whether the loop has humans in it who can climb to the next hill, or whether it ships a metric and calls the climb someone else's problem.
The watch item for the next year is whether the AI-native companies that raised in 2024 and 2025 publish org charts at all, or whether they publish metric maps. The wire will report that Acharya said "loops" are replacing org charts. The harder question is which companies reorganize around the climb between local maxima and which bolt an agent onto the building they already have.