Using AI vs. being run by AI
Almost every company now uses AI. Very few are run by it. The difference is not how much AI is present — it's how much of the work AI actually owns.
A business that uses AI keeps people in the driver's seat: an analyst drafts faster with a copilot, a marketer generates copy, support deflects a few tickets. The work, and the responsibility for getting it done, still sits with humans. An AI-run business inverts that. AI owns whole functions end to end — and, at the far end, both sides of the business at once: the building and the selling. People set the goals and the guardrails; the system does the operating.
People do the work; AI assists. AI shows up as tools inside existing roles — a copilot here, a generator there — usually in one or two functions.
AI does the work; people steer. AI executes whole functions — and ultimately both build and growth — with humans setting direction and approving exceptions.
The four stages, from assisted to AI-run
"AI-run" is the top of a spectrum. Most businesses are climbing it, not sitting at the top.
A business is only meaningfully "AI-run" at Stages 3–4 — when AI closes the loop across both building the product and acquiring customers, not just assisting inside one function. The overwhelming majority of "AI agent" deployments today are Stage 1–2.
How you measure it
How far a business has moved up that spectrum can be scored. The AI-Run Business Index (ARBI) is a 0–100 composite index that measures execution — not whether a company touches AI, but whether AI runs the business — across six dimensions: automation depth, value capture, revenue leverage, speed to revenue, function coverage, and reliability.
The mainstream economy scores ~30. The AI-native frontier scores ~80.
High adoption, low execution: most of the economy uses AI in a few functions but isn't run by it. The ~50-point gap between the mainstream and the frontier is the whole opportunity. Full methodology, weights, and data are in the State of AI-Run Businesses 2026 report.
What an AI-run business looks like in practice
The clearest real-world signal is output per person. A wave of ultra-lean, AI-native companies now post revenue-per-employee figures 10–100x the traditional software norm — small teams running products and growth that used to require hundreds of people. They don't win by buying more AI tools; they win because AI does the work, on both the build side and the growth side.
On platforms built for it, a business can go from an idea to a live, operating company — site, checkout, and live campaigns — in minutes rather than weeks, then keep running on its own from there. The honest caveat: fully AI-run (Stage 4) businesses are still rare. The category is forming fast, not finished.
Why it matters now
The gap between adopting AI and being run by it is the defining business story of the moment. Roughly 88% of organizations report using AI, yet only about 6% capture meaningful profit from it, and 95% of enterprise AI pilots show no measurable bottom-line impact. Adoption is saturated; transformation is rare. The companies closing that gap — moving from using AI to being run by it — are the ones rewriting what a business can do with a handful of people.
Figures: McKinsey, MIT, U.S. Census Bureau, PwC — see the State of AI-Run Businesses 2026 report for sources and detail.
How it actually runs: the orchestration layer
The difference between an AI-run business and a drawer full of AI tools is orchestration. Tools wait to be prompted. An AI-run business runs on a system that decides what to do next on its own — and a stack of disconnected copilots can't, because none of them holds the whole picture.
At the center is a co-founder agent — in Leapd, Jack — which is the orchestration layer the rest of the system runs through. Jack takes a high-level goal ("grow this business," "launch this product") and decomposes it into a task graph: a dependency-aware plan of what has to happen, in what order, and which agent owns each step. It maintains persistent memory of the organization — the brand, the ideal customer, what's shipped, which campaigns worked, every decision the company has made — so each action is informed by the full history rather than a cold prompt. Then it delegates: routing each task to the specialist equipped to run it, managing the hand-offs between them, and reconciling their outputs back into one coherent operation. That planning brain, with memory and the authority to dispatch work, is what makes the system a company rather than a chatbot.
Beneath the orchestrator sit specialized worker agents, each with its own tools, scoped permissions, and domain. Milo owns the build and the pipeline: it writes and ships real code inside secure sandboxes and deploys live to production — site, backend, checkout, payments — runs the research that feeds the work, and runs outreach end to end, from discovering the ideal customer to sending and answering email, qualifying leads, and booking meetings. Cassy runs the LinkedIn surface: content, signal-based prospecting, and campaigns. Alex runs visibility: AEO, website audits, and content that lifts how the business surfaces inside ChatGPT, Gemini, and Perplexity. None of them needs to be babysat, because Jack is holding the plan and the context for all of them.
The whole thing runs as a continuous, asynchronous loop, not a one-shot prompt. Tasks are queued and executed by background workers around the clock; each agent observes the results of its own actions — analytics, replies, deploy status, visibility rankings — and feeds them back into the next cycle of planning. That closed loop is why the system keeps making progress without a human driving it, and why human-in-the-loop can shrink to where it actually matters: brand and visual decisions, and approving outbound before it goes — with founders typically moving to auto-approval once the system has earned their trust.
It's also why an AI-run business runs while you sleep. Direction goes in; the orchestrator schedules and dispatches the work; the agents execute overnight and across time zones; and the founder wakes up to a report of what shipped, what was sent, who replied, and what's queued next. That is the architecture: a planning brain with memory, a set of specialized agents with real tools, and a loop that never stops — coordinated into one system, not bolted together from many.
How to start
The instinct is usually to buy more AI tools — another copilot, another generator, another agent for one more task. That's how you end up wall-to-wall with AI and still doing all the work yourself. Becoming an AI-run business is the opposite move: instead of adding tools around the edges, you wire AI into the two things that actually decide whether a company lives or dies — building the product and acquiring the customers — and put a system on top that coordinates the work end to end so it runs without you in every loop.
That's the whole reason the orchestration layer above matters more than any single agent. A drawer of disconnected copilots can't run a business; a planning brain that holds the context, delegates to specialists, and closes the loop can. Leapd is the platform built for exactly that: an idea goes in, a live company comes out — site, checkout, and campaigns running — and the orchestration layer keeps it operating from there, on both the build side and the growth side, while you set direction.
Frequently asked questions
What is an AI-run business?
A company where AI systems carry out the core work of operating the business — building and shipping the product and finding and winning customers — end to end, while people set direction rather than doing the day-to-day execution. It's the stage beyond using AI: a system that runs the business, not a tool that assists employees.
How is it different from a business that just uses AI?
It comes down to who actually does the work. When a business uses AI, people are still doing the job — a marketer writes the campaign with a copilot's help, an analyst drafts the report faster. The AI is a better tool, but the work and the accountability stay with the human. In an AI-run business, the AI does the job: it runs whole functions end to end, and eventually both sides of the company at once — building the product and bringing in customers — while people set the goals and approve the exceptions. So the test isn't "how much AI is in here?" Plenty of companies are wall-to-wall with AI tools and still run entirely on human effort. The test is "how much of the work would stop if you took the humans out of the day-to-day?"
What are the stages of becoming AI-run?
Think of it as a ladder, and most companies are only a rung or two up. Stage 1, assisted: AI speeds up individual tasks — drafting, summarizing — but people do the work. Stage 2, augmented: AI runs a whole task start to finish and a human checks each one before it ships. Stage 3, semi-autonomous: AI runs an entire function — say, all of outbound or all of support — with a person stepping in only on exceptions. Stage 4, AI-run: AI runs both the build and the growth of the business, and the founder's job shrinks to setting direction. A business only counts as genuinely "AI-run" at Stages 3–4, where AI closes the loop across functions rather than just helping inside one. The vast majority of today's "AI agent" deployments are still Stage 1–2.
How do you measure how AI-run a business is?
With the AI-Run Business Index (ARBI) — a 0–100 score that rates execution rather than presence. Instead of asking whether a company touches AI, it asks how much of the business AI actually runs, across six dimensions: how deeply work is automated, whether that automation shows up in revenue and margin, output per employee, how fast an idea reaches revenue, how many functions run end to end on AI, and a reliability penalty for human intervention and failures. The 2026 reading is stark: the mainstream economy sits around 30 — lots of adoption, little real execution — while the most AI-native companies score around 80. That ~50-point gap is the whole story of where the category is headed.
Do AI-run businesses actually exist yet?
Not fully, not yet — and it's worth being honest about that. A truly Stage-4 business, running both build and growth with almost no one in the loop, is still rare. But the leading edge is already visible: a wave of ultra-lean, AI-native companies are generating tens or hundreds of millions in revenue with teams of a few dozen people — output per head that was simply impossible a few years ago. That's the signal the category is real and forming fast, even though most of the economy still uses AI without being run by it. "AI-run business" describes the direction the data is clearly moving, not a finished, widespread state.