Loading playbook…
ALL-IN · EXTRACTED
7 tactics for building genuinely hard hardware, told through the design of a robot's hand.
"I believe Optimus is going to be the greatest product ever created by humanity." — Elon Musk
Strip away the Mars philosophy and the Starlink-phone teaser and you get a clinic in how to build something physically hard: what to solve first, when to make your own parts, and how to find the real bottleneck instead of the average one.
On the humanoid robot, he says "the hands inclusive of the forearm are a majority of the engineering difficulty of the entire robot." A human hand has 27 to 28 degrees of freedom. You cannot have a general-purpose robot without solving it, so it gets solved first.
THE PLAY
Find the single hardest core component your product depends on and solve it first. Everything else is downstream.
There is no supply chain for humanoid-robot actuators. He says they could not buy them for any amount of money, so they designed every motor, gearbox, and controller "from physics first principles."
THE PLAY
When no supplier exists, do not wait for the market. Build the part from physics first principles.
He chose the human form on purpose. Humans designed the world for humans, so a humanoid robot "will be immediately backwards compatible with what we've built the world for." The shape is a compatibility decision, not an aesthetic one.
THE PLAY
Design your product to fit the environment it enters, so adoption does not require the world to change first.
Tesla's AI hardware and software teams co-design the chip, so they know exactly where the limits are. The next chip is roughly 8x more compute, but because they also fixed a specific suboptimal operation that ran in about 40 emulated steps, the real-world gain on that limit is around 40x.
THE PLAY
Co-design adjacent layers so you can attack the actual bottleneck rather than improving the average.
On scaling, his rough rule is that 10x more compute roughly doubles the intelligence, a logarithmic curve. Diminishing, yes, but going from 100 to 200 is still a very big deal.
THE PLAY
Do not abandon a lever just because returns diminish. Sometimes 10x of input for 2x of output is still the best move you have.
His real test for a Mars city is whether it survives if the ships from Earth stop coming. It is a clean redundancy mental model: a system is only truly resilient if it keeps running after its lifeline is severed.
THE PLAY
For anything critical, ask whether it survives if the supply line is cut. Build to pass that test.
For the next model, instead of just feeding in more data, they use heavy inference compute to review each piece, removing falsehoods and adding the missing context. Better inputs, not just more of them.
THE PLAY
Before scaling up the volume of data or inputs, invest in correcting and enriching what you already feed the system.
YOUR ACTION PLAN
All the plays, back to back. Use this as your checklist.
Solve The Hardest Core Component First
Identify and solve the single hardest core component first.
Build From First Principles When No Supplier Exists
Build from first principles when no supplier exists.
Make It Backwards Compatible With Its Environment
Design for compatibility with the environment your product enters.
Co-Design Layers To Attack The Real Bottleneck
Co-design layers to hit the real bottleneck, not the average.
Know That 10x Of Input For 2x Of Output Can Still Be Worth It
Keep pulling a lever even when returns are logarithmic, if the absolute gain is large.
Design For The Resupply-Cut Test
Stress-test critical systems against having their supply line cut.
Clean Your Inputs Before Adding More
Clean and enrich your inputs before adding more of them.
Ep. 001
8 tactics for building companies and not dying, from a guy who kept two of them alive on the edge of bankruptcy.
Read playbook →Newsletter
One email per drop. No spam. Unsubscribe anytime.
ALL-IN · EXTRACTED BY PODEX