Run with, not through is the difference between AI the team carries and AI that routes through one person. Run through means every important decision, check, and handoff waits on a single individual, usually the founder. Run with means the judgment lives in a shared system the team executes against, so the work moves without one person in the middle of every step.

It is worth being precise, because the two can look similar from the outside. A business running AI through the founder and a business running AI with the team can own the same tools and show the same usage numbers. The difference is the wiring underneath. In the first, the founder is the strategist, the quality check, the escalation point, and the place decisions wait, and adding AI just means the AI waits there too. In the second, the context and the standards sit in a place the team can reach, so a job gets run against them without routing back to one person. This is the operating distinction the whole idea of a team running with the founder rests on, and it is the one phrase this argument is built around.

Because routing the work through one person throttles it at that person, and the person at the top is usually the lightest AI user in the building. That is the uncomfortable irony of the through model: the node everything depends on is the node least likely to actually be using the tool.

The data is blunt about who the lightest user is. Across surveys of senior leaders, roughly 70 percent of senior executives use AI less than an hour a week, including about 28 percent who never use it at all, even while many of them mandate AI for the people below them. When the work routes through the person using AI least, adoption is capped at the weakest node by design. And the through model fails a second way, through bad information: leaders consistently misjudge their teams, so the router is not only the lightest user but also the worst-informed about how the work is actually landing.

AI run through the founder stalls at the founder, who is usually the lightest user in the company.

The contrast with running with the team is now an experimental finding, not a slogan. A pre-registered field experiment of 776 professionals found that individuals working with AI matched the performance of two-person teams working without it, and that AI broke down the silos between functions, pulling specialists who normally stayed in their lane toward shared, balanced solutions. AI did not just make a person faster. It changed the unit of work, so a job that used to need a particular person in the middle could be carried more broadly. That is what “with” delivers and “through” cannot: the work moves at the level of the work, not at the level of one calendar.

You move the judgment out of your head and into a system the team can run, and you back it with support rather than mandate. Running with the team is not stepping back and hoping. It is making the context, the plays, and the standards available where the team works, so they can execute without waiting on you, and actively supporting that use rather than requiring it.

Manager support, not the mandate, is the lever that turns scattered individual use into how a team operates. Employees who strongly agree their manager actively supports team AI use are far more likely to use AI weekly, to find the tools useful, and to say AI lets them do what they do best, yet only about 28 percent strongly agree their manager actually supports it. That gap is the founder-bottleneck wearing different clothes. When everything runs through the founder, the founder cannot be the supportive backstop the team needs, because they are the constraint the support has to route around. The diagnosis of how the founder became the bottleneck in the first place is its own piece; the point here is that you cannot support what you are simultaneously blocking.

Employees whose manager actively supports team AI use are far more likely to use it weekly, yet only about 28 percent say their manager does.
Gallup, 2025

You stop being the place answers come from and become the place the standards are set, once. The shift from answer to architect is the entire job reframe of a founder who wants the business to run with them: your value moves from being in every decision to having built the system every decision runs against. This is what a good entrepreneur’s actual job turns out to be.

The reason this is hard is that most AI does not help you make the move, it deepens the hole. Add a tool the founder operates and you have added one more thing routed through the founder. The honest research bears this out: a vendor study found that 96 percent of organizations report isolated AI productivity gains that have not become company-wide transformation, with the small minority that succeed using AI collaboratively rather than in silos. Isolated gains are the through model’s signature. They show up as one person getting faster, never as the business changing how it runs.

Here is where the show beats the tell. The pain is that the judgment lives in your head, so every job waits on you and the AI just queues behind you with everyone else. The Works mechanism that relieves it is Expert-Grade Workflows, plays built on real operating methods that hold the standard so the team runs the whole job against it, paired with the autonomy controls (Copilot, Pilot, Autopilot) that let you set how much runs without sign-off, per workflow. The gain is a business that executes to your standard without routing through you. The affordability that makes this real for a founder-led business is the $49 Pro tier, capability without the senior hire. And it is not theory for us: run with six teams, with revenue per employee landing two to three times higher. The fuller account of how that played out is its own story, and the part that matters here is the wiring: it only worked once the AI stopped living on one login and started running with the team. The failure mode when that shift does not happen, the team going quiet rather than telling you it routes through you, is the polite silence.

Run the business with you, not through you. Get early access. Or start with the Bottleneck teardown to find where the work is still routing through one person.

What changes when decisions are made without me?

Speed uncouples from your calendar, and quality does not drop because the standard moved into the system rather than leaving with you. The NBER field experiment gives a concrete measure of the ceiling that lifts: individuals working with AI matched the output of two-person teams working without it. That multiplier only materializes when the AI is running with the team against a shared standard, not waiting in a queue behind the founder for every step.

Does my support for AI actually change whether my team uses it?

Yes, more than almost any other factor. Employees who strongly agree their manager actively supports team AI use are dramatically more likely to use it weekly, find it useful, and say it lets them do their best work. Support, not a mandate, is the lever. The catch is that a founder who routes everything through themselves cannot be that supportive backstop, because they are the constraint the team is working around.

Is running through me not safer, since I guarantee the quality?

It guarantees a bottleneck at the lightest user in the building, which is the opposite of safe. Quality at the speed of one calendar is not quality the business can rely on. The durable version comes from setting the standard once in a system the team runs against, backed by your support, which Gallup data ties to two-to-eight-times higher adoption. You get more reliable quality, not less, by stopping being the single point it routes through.

Why does AI run through one person fail even when that person is highly capable?

Routing is the constraint, not capability: a highly capable person is still one person, and the survey data shows that person at the top is typically the lightest AI user in the building, with roughly 70 percent of senior executives using AI less than an hour a week. Talent concentrated in a single node does not scale the work; it throttles it. Moving the AI to run with the team rather than through that node is what lets the multiplier land across the operation instead of sitting behind one calendar.

How is running with my team different from just delegating more?

Delegation hands someone the task. Running with the team puts the judgment that drives the task into a system the whole team executes against, so the work does not bounce back when an edge case appears. The difference is that delegation still routes the exceptions back to you, while running with means the system holds the standard and the escalation line is explicit. The mechanics of encoding that judgment are in the five questions your team asks you every week.

1. Gallup, Manager Support Drives Employee AI Adoption, 2025
2. NBER Working Paper w33641, The Cybernetic Teammate (Dell’Acqua, Sadun, Mollick et al.), 2025
3. Fortune, CEOs are mandating that employees use AI. They’re hardly using it themselves, 2026
4. Atlassian Teamwork Lab, AI Collaboration Report 2025, 2025
   
   

Keep reading:

• The founder bottleneck
• A good entrepreneur’s job
• The polite silence
• Team Runs With Me