The Right Level of AI
Most leaders are stuck in one of two AI conversations. Zoom out and there are three levels — and the right target is somewhere in between.
Most AI conversations are stuck in one of two places.
Some leaders are focused on tools that make their staff faster — better autocomplete, smarter assistants, an AI copilot for every workflow. Others are focused on agentic AI — autonomous systems that operate without supervision. Both conversations are happening. Neither is the full picture.
Zoom out, and these are not two competing visions. They are two points on a spectrum that has at least three distinct levels. Most leaders cannot tell you which level their organization is operating at, which level it should be targeting, or what has to change to get there. That is the gap this article is trying to close.
There are three distinct levels at which an organization or individual can engage with AI. Each requires a different operating model. Each demands a different mix of judgment, oversight, and accountability. And the right target for most organizations right now is somewhere in between — call it Level 2.5. Positioning for it requires organizational design changes most leaders have not started making.
The three levels
Level 1 — AI as accelerator. You are still the doer. AI is the power tool that makes you faster — the air nailer instead of the hammer. Every meaningful decision passes through your hands. Two equally skilled employees, one using AI and one not, will produce different output at different speeds. The one with AI is faster, sharper, more thorough. This is where most professionals are operating right now.
Level 2 — AI as delegate. You become the director, not the doer. You send AI off on missions, then review and verify what comes back. Fewer touch points per task. More time spent setting direction and inspecting output. You are still accountable, but the operating model has fundamentally changed. You no longer make things happen with your hands. You make things happen through delegation.
Level 3 — AI as virtual organization. Entire functions are agent-based. Specialist agents talk to other specialist agents. A virtual head of finance coordinates with a virtual head of operations. You provide top-level direction; the system figures out the work between roles. This is the more extreme case — visible on the horizon, not yet operational at scale.
Where most organizations actually are
Level 1 with Level 2 ambitions.
Leaders are watching their teams use AI to draft emails, summarize documents, and accelerate analysis — Level 1 work. Then they look at the agent demos circulating in industry conversations and assume Level 3 is around the corner. The gap they miss is the entire middle layer — the operating model, the verification discipline, the organizational design changes that Level 2 actually requires.
Most organizations have not made the shift from "amplifying my work" to "directing work done on my behalf." Until they do, Level 3 conversations are theater.
The Level 2.5 target
In private business, Level 3 is plausible. Accountability can be structured around outcomes, and shareholders accept the trade-off when the math works.
In regulated, accountability-bound environments — healthcare, financial services, anything with a personal-accountability tail — Level 3 hits a ceiling that is not technical. Decisions that affect patients, consumers, or the public have a human name attached to them. Personal accountability does not scale with AI autonomy. An organization cannot say "the virtual CFO approved it" when something goes wrong.
The practical position is Level 2.5. Far enough into delegation to capture the operational gains. Not so far that the human accountability tail breaks. This is not a technology limitation. It is a governance limitation. Leaders who assume Level 3 is universally achievable are not seeing the constraint clearly.
Why this matters now
Most organizations are designed around humans and large staffs. If salary is your biggest line item today, it may shift toward token spend tomorrow as you move into Level 2.
Here is the leadership reframe most are missing. Even if headcount decreases as token spend grows, the value of the work being done is roughly the same. The money is moving from one column to another. It is not necessarily going down.
Leaders who pitch AI adoption as a cost-reduction story are setting themselves up for credibility loss. The real story is value capture and capability scale — doing more with the same budget, doing it faster, or doing things that were previously impossible. Cost reduction is sometimes a side effect. It is not the headline benefit, and treating it as one will burn political capital when the savings do not materialize on the timeline leadership expected.
The leadership frame
You do not need to predict whether Level 3 arrives in two years or ten. You need to know which level your organization is actually operating at, which level it should be positioning for, and what has to change to get there.
Most leaders are answering a question they do not yet have the right framework for. The levels framework is the framework. Use it.
The technology is moving. The accountability model is not. Position for the gap.