For a startup founder, deciding what to work on is a challenging but rewarding task that lays the foundation for everything ahead. How does one know what to work on, or whether it is the right thing? Well, nobody knows the exact answers but there are roughly two approaches you could take to look at this question from different angle: Goal and idea.
A goal is more powerful than an idea for finding knowledge
Discovering and creating knowledge requires a goal-driven approach: setting an objective, formulating hypotheses, defining constraints, running experiments, and evaluating the results until it is proven true. Knowledge must survive repeated testing by clearly explaining what and why under set of constraints.
A goal-driven approach does not care which method is used as long as it solves the problem. This opens a larger search space and provides greater flexibility to explore and test different methods within the imposed constraints. Working towards a goal will make you ask questions no one is asking, tackle problems every one is overlooking, and will give you unique perspective that’s differentiated from the rest of the market.
Most important of all, finding truth takes time. It takes an incredible amount of effort and patience to move from asking the initial question to uncovering an answer and establishing it as “truth.” Working toward a goal is like following a lighthouse. You know exactly where to go, even if the path isn’t a straight line. A goal provides the persistence, patience, and motivation that makes much easier sit on and work on it for much longer.
An idea is more powerful than a goal for capturing opportunities from disruption
An idea-driven approach is more opportunistic. If a goal-driven approach is scientific, an idea-driven approach takes a more naive approach: it begins without a fixed objective and follows curiosity and taste instead. The opportunity is often created by new technology or trend in the market and requires founders to notice the emerging gaps in the market.
It is better at “getting lucky”. Idea usually starts from one’s own interest and follows the taste much more than the goal. Such characteristic could seem risky (which it is), but also allows faster exploration and execution, expanding one’s surface area of luck. A lot of companies have had success this way (Airbnb, Stripe etc.) and finding what to work on by working is still one of the more effective ways to build a startup from ground up.
It also serves a different purpose from a goal-driven approach. Its primary purpose is usually not to discover empirical truth but to recognize what has become newly possible and turn existing knowledge into a product. Facebook, Airbnb, and Uber are great examples. Companies taking the idea-driven approach therefore tend to operate mostly at the application layer on top of existing technology.
How to choose
They are not mutually exclusive. Goals can emerge from an idea, and ideas can emerge from pursuing a goal. The important question is which one you choose to organize your team around. The idea of building the best physics simulation engine is genuinely cool, but why? On the other hand, solving the sim-to-real gap to make robotics engineering more reliable has a fascinating purpose that one can spend years pursuing.
There is no single way to build a startup. Ultimately, it comes down to the kind of company you want to build and the product you want to create. However, for those who aim to push the boundaries of frontier technology, the goal-driven approach will be much more powerful.
Towards RSI
The goal-driven approach has been central to the startup I am building. My goal is to discover new knowledge through research and invent novel approaches that fundamentally change how robots are built. The objective is clear: build a recursive self-intelligence system for robotics that enables the autonomous optimization of physical AI, bounded only by the laws of physics. As delusional and far-fetched as this may sound, I predict it will become reality within ten years. The first blocker is building a closed loop that can propose changes to a robot’s design or control, evaluate them in simulation, validate them on physical hardware, and learn from the results. From there, the work is to clear each remaining blocker, one by one, until we reach the goal.
More coming soon.