Vitalik Buterin's 'AI as the engine, humans as the steering wheel' wasn't about productivity — but it should have been. Here's the framework we're building Exponential around.
In February 2025, Vitalik Buterin published an essay called AI as the engine, humans as the steering wheel. His argument was about governance — how to design systems where AI does the heavy computational work while humans provide a small amount of very high-quality directional input. Not micromanagement. Not abdication. A division of labor based on what each side is actually good at.
He wrote:
Humans provide only a small amount of information into the system, perhaps only a few hundred bits, but each of those bits is a well-considered and very high-quality bit. AI treats this data as an "objective function", and tirelessly makes a very large number of decisions doing a best-effort at fitting these objectives.
Vitalik was talking about DAOs, prediction markets, and funding mechanisms. But reading it, I couldn't shake the feeling that he'd also described — maybe accidentally — the future of personal productivity.
Because this is exactly what's broken about how most people manage their work. And it's what we're building at Exponential.
The Two Failure Modes
Most people oscillate between two states:
Autopilot. You wake up, open your inbox, react to whatever's loudest, and by 6pm you've been busy all day but moved nothing important forward. You're the engine and the steering wheel, and neither is working well.
Rigid planning. You spend Sunday night building an elaborate system — color-coded calendars, prioritized task lists, weekly reviews. It works for about two weeks. Then one unexpected meeting derails Monday, and by Wednesday the whole thing has collapsed. You're steering so hard you forgot to drive.
There's a third option. And it maps almost perfectly onto what Vitalik describes.
The Self-Steering Method
The idea is simple: you steer, AI drives.
You set the direction — a few high-quality decisions about where you're going. The AI agent handles the thousands of micro-decisions about what to do next, when, and in what order. You course-correct together.
This isn't a to-do list app with GPT bolted on. It's a fundamentally different relationship between a human and their productivity system.
It has four phases.
1. Set Your Coordinates
Not "goals" in the corporate OKR sense. More like: what does a good quarter look like? What are the 2-3 things that, if they happened, would make everything else easier or irrelevant?
These might be:
- Launch publicly at a specific event
- Close a key partnership
- Get 50 real users actively engaging with the product
These are directional. Vitalik would call them your "objective function" — a small number of bits, but high-quality ones. The agent needs to understand where you're going before it can make any useful suggestion about what to do today.
Most productivity systems skip this step entirely. They start at the task level. That's like optimizing the engine before deciding where you're driving.
2. Triage Ruthlessly
Every week, your coordinates generate competing demands. Emails arrive. Deadlines shift. New opportunities appear. The old approach: you manually sort through everything, usually by anxiety level.
The self-steering move: the agent surfaces what's most leveraged right now — based on your coordinates, deadlines, calendar, energy, and what's actually moving.
The manual version of this is a single question: "If I can only do 3 things today that move my coordinates forward, what are they?"
That's it. The agent automates that question. And unlike you at 8am with coffee, it doesn't have recency bias, doesn't confuse urgency with importance, and doesn't forget about the thing you've been avoiding for two weeks.
This is the Vitalik pattern in miniature: the mechanism (your priorities) is simple and human-directed. The execution (figuring out what's most leveraged given a hundred variables) is where AI excels.
3. Execute in Blocks
Not a to-do list. A short list generated from your priorities, matched to the time you actually have. The agent knows your calendar. It knows what's urgent. It knows what you keep pushing to tomorrow.
The human version: block 90 minutes for your #1 priority before you open email. Every day. This works, and you should do it even without an app.
The agent version: your day gets built around what matters, not what's loudest. The engine runs. You steer.
4. Reflect and Recalibrate
This is what makes it self-steering and not just another planner.
At the end of the day or week, the loop closes: did what I worked on actually move my coordinates? What surprised me? What should shift?
The agent learns from this. Your priorities get sharper over time. The system steers with you, not just for you.
Vitalik makes a key distinction in his essay: he argues against putting a single AI in charge. Instead, he favors systems where "the mechanism is simple, and the intelligence comes from participants operating at the edge." The Self-Steering Method works the same way. The framework is dead simple — coordinates, triage, execute, reflect. The intelligence comes from the feedback loop between you and the agent, sharpening over time.
Why This Isn't Just Productivity Advice
Here's the thing: you can run the Self-Steering Method with a notebook. Set coordinates quarterly. Ask the triage question daily. Block time for deep work. Reflect weekly. It works.
But the unlock — the thing that turns a good habit into a compounding advantage — is when an AI agent runs the loop with you. Because:
- You forget. The agent doesn't.
- You have recency bias. The agent weights by your coordinates.
- You avoid hard tasks. The agent notices the pattern and surfaces them anyway.
- You can't hold 50 variables in your head. The agent can reason across your entire calendar, project list, and priority stack simultaneously.
This is the same asymmetry Vitalik identifies in governance: humans are better at deciding what matters (values, direction, priorities). AI is better at computing how to get there (optimization, scheduling, pattern recognition across large data sets).
Neither is sufficient alone. An AI agent with no human direction optimizes for metrics that might not matter. A human with no AI support drowns in complexity. Together, you get something that compounds.
The Car Metaphor (Which Is Actually Pretty Good)
Vitalik's metaphor is worth taking literally for a moment.
When you drive a car, you don't control each piston firing or calculate the optimal fuel-air mixture. The engine handles that — thousands of micro-decisions per second. You provide the direction: turn left here, speed up, take the highway instead of surface streets.
Your input is low-bandwidth but high-judgment. The engine's output is high-bandwidth but zero-judgment.
Now imagine driving a car where you had to manually manage the engine and navigate. That's most people's relationship with their productivity system. They're trying to be the engine and the steering wheel simultaneously, and burning out doing both badly.
Self-steering means accepting the division of labor. You decide where to go. The AI figures out the best way to get there. And when the route isn't working, you adjust — not by micromanaging the engine, but by pointing in a new direction.
What We're Building
At Exponential, the Self-Steering Method isn't a blog post framework — it's the product architecture.
- Coordinates are your goals and north stars in the app
- Triage is the AI agent that reasons about priority across your entire stack — not just displaying tasks, but ranking them by leverage against your coordinates
- Execution is daily plans built from your priorities and real calendar, not wishful thinking
- Reflection is the feedback loop where the agent learns what actually worked and adjusts
We're building the engine. You bring the steering wheel.
Stop managing. Start steering.
Exponential gives you an AI copilot that runs your productivity system — so you stay focused on the work that matters.