Innovation comes in two kinds: Bottom-up or top-down.

Bottom-up

You start with tinkering, small hacks and serendipity.

Examples:

Drugs like Insulin and Penicillin were discovered by coincidence.
The Transformer architecture allegedly was discovered by throwing a bunch of "tricks" (eg. positional encodings) at Bahdanau's RNN minus the recurrence.
ChatGPT, according to Sam Altman: “It took us four and half years to launch a product. [...] We were building a technology without any idea of who our customers were going to be or what they were going to use it for.”
Paul Graham argues that great startup ideas often emerge from working on things that could be dismissed as toys.
Why Greatness Cannot Be Planned challenges goal-first marching because it can block exploration.

Upside: real, often non-consensus, signals from the world.

Downside: most paths lead nowhere, you need lots of shots on goal; cultures that make experiments cheap help.

Top-down

You start with a vision about the future, then design product and tech to make that vision true.

Examples:

Nation-scale mission programs like the U.S. Interstate Highway System or the Manhattan Project were centralized, single-objective efforts. A lot of this is now happening in China, eg., with power stations or water diversion.
SpaceX's Vision: make life multiplanetary. Reusability, Starship, and even Falcon-era steps are milestones on a long-term Mars roadmap.
Palantir "floundered for years, barely getting any real traction in the gov space, and doing the opposite of the ‘lean startup’ thing" (Source)
Peter Thiel's Zero to One argues for plan-first founders: “You could build the best version of an app that lets people order toilet paper from their iPhone. But iteration without a bold plan won’t take you from 0 to 1.”

Upside: clarity, focused execution, and team alignment; easier to coordinate capital and GTM around a single thesis.

Downside: rigidity and blind spots; if the thesis is wrong, commitment amplifies the error. Course corrections get costlier the further you go.

Bottom-up experience gives you top-down vision

People who have strong top-down visions tend to have bottom-up scars. The cuts teach taste.

In research, folks who have tinkered a lot of with ideas, eg., by breaking existing methods, formulate and frame better hypotheses because their intuitions inform them about what really matters.

In startups, someone who built a product and made even just $1,000, gained tacit knowledge about product and go-to-market. That's why investors love to invest in 2nd time founders.

Still, bottom-up can win on its own; chance and stubborn work often do.