ExpertsFMWhy we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)
Lenny's Podcast: Product | Career | Growth · May 17, 2026
5 Selected Lessons:
framework · 3 reasoning_pattern · 1 priority · 1
Lessons from legendary builders: ambition, uncompromising bar, clean decision velocity
Source
Lenny's Podcast: Product | Career | Growth — Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple) (May 17, 2026) · at 1:23:45 ▸
When this applies
when identifying which specific leadership behaviors most improve organizational output, as distinct behaviors rather than a single leadership style
The takeaway
Distinguish the specific mechanism by which a leader's behavior improves outcomes (bigger ambition, higher quality floor, faster correct decisions) so you can deliberately adopt the specific practice rather than vaguely imitate 'great leadership.'
Why it works
Each trait operates through a different causal channel: scale-of-ambition framing (Altman) expands the solution space considered; an unwavering excellence bar (Jobs) raises the floor of acceptable output across the org; and decentralized-but-reviewed decision rights plus leadership technical fluency (Zuckerberg/Bosworth) preserves both speed and quality at scale.
One level up
Effective leadership is a bundle of separable, learnable behaviors, not a single personality trait to imitate wholesale.
Where it stops holding
This stops working when a leader applies these traits without organizational infrastructure to support them (e.g., pushing ambition without resources, enforcing a quality bar without giving people tools to meet it, or decentralizing decisions without competent reviewers).
Detail
hierarchy:
- Sam Altman: push ambition — ask 'why not 100x or 10,000x' to expand what's considered possible
- Steve Jobs: hold an unwavering excellence/technical bar across the company regardless of comfort
- Mark Zuckerberg/Andrew Bosworth: push decisions to the lowest competent level, make decision processes explicit, and have leadership stay technically fluent enough to engage deeply with real tradeoffs
rationale: Being told something isn't good enough by a leader with a non-wavering bar is 'extremely motivating' even though it doesn't feel good in the moment; decentralized-but-reviewed decisions with technically engaged leadership let a fast-growing company move quickly without losing rigor.
In their words
Sam is really good at saying, why not more? Why not 100x or 10,000 X? you're thinking too small.
Robotics/hardware supply chain layering
Source
Lenny's Podcast: Product | Career | Growth — Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple) (May 17, 2026) · at 15:08 ▸ · at 17:35 ▸ · at 20:21 ▸
When this applies
when mapping dependency risk in any multi-tier supply chain for a physical product
The takeaway
Decompose a supply chain into its raw-material, processing, integration, and assembly layers, and identify which layers lack geographic or source redundancy — those are your true points of fragility.
Why it works
Each layer (raw material -> processing -> component integration -> subassembly -> final product) has been outsourced to specific geographies over decades, so risk concentrates wherever a layer has no domestic/alternative alternative.
One level up
Resilience requires understanding dependency structure at every layer, not just visible final-assembly risk.
Where it stops holding
This stops working when a company doesn't have visibility into upstream layers (e.g., sub-tier suppliers' suppliers) or when switching costs across layers are low enough that disruption is easily absorbed.
Detail
procedure:
- Start with raw materials (e.g., magnets)
- Identify who can process them
- Identify who can integrate them into components (e.g., actuators)
- Identify who can integrate components into subassemblies/final product
- Assess which layers are foundational (backed by material science breakthroughs like batteries, actuators) vs. less critical (e.g., die-cast parts)
benchmarks: {"foundational_layers":"magnets/materials, actuators, batteries","less_critical_layers":"die-cast/machined parts"}
rationale: Each layer of the chain has essentially been outsourced over 25 years to China, Japan, Korea for scale and cost; true independence requires rebuilding capability at every layer, not just final assembly.
In their words
And each layer of this chain has essentially been outsourced over the last 25 years to countries like China, like Japan, like Korea.
Apple's 'back of the cabinet' first-principles design discipline
Source
Lenny's Podcast: Product | Career | Growth — Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple) (May 17, 2026) · at 26:50 ▸
When this applies
when building any product and deciding whether to extend quality discipline to parts customers will never directly observe
The takeaway
Apply your quality bar uniformly, including to invisible parts of the product, because the discipline of justifying every decision — not the decision itself — is what produces organizational excellence.
Why it works
Forcing a team to justify even invisible design decisions against first principles ('why are we building it this way, what outcome matters') trains the organization to consistently identify what truly matters, which then surfaces as simplicity in the final product.
One level up
Process discipline applied even where no one is watching compounds into talent and product excellence that's hard to copy.
Where it stops holding
This stops working when resource constraints make it genuinely wasteful to polish unseen details relative to higher-leverage work, or in very early-stage prototyping where speed matters more than polish.
Detail
procedure:
- Ask why you're building it this specific way for every component, visible or not
- Hold the excellence bar constant regardless of visibility
- Let what truly matters naturally surface through this rigor
rationale: Apple's dedication to finishing even the inside/back of a device (the 'back of the cabinet' lesson from the cabinet maker Jobs admired) forced engineering, ID, and ops to think about core purpose on every decision, producing people who now hold key positions across the industry.
In their words
And that goes very, very deep at Apple, where every single design decision, even on the inside of the device, is considered.
Four-part hardware execution playbook
Source
Lenny's Podcast: Product | Career | Growth — Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple) (May 17, 2026) · at 33:07 ▸ · at 35:10 ▸ · at 37:27 ▸
When this applies
when executing any project with few irreversible checkpoints and strong deadline pressure
The takeaway
In any irreversible, cycle-limited execution process: fix goals early, attack the highest-risk unknowns first, concentrate iteration on what the user actually touches most, and act immediately on anything known rather than banking time.
Why it works
Hardware's irreversibility and long iteration cycles mean early decisions compound: unclear goals waste iterations, easy parts first hide late-stage risk, under-iterated touchpoints hurt the core user experience, and delay directly eats your fixed number of design cycles.
One level up
Treat scarce iteration cycles as the real currency of execution, not calendar time.
Where it stops holding
This stops working when the project is cheaply reversible (software with instant redeploys) or genuinely open-ended with no shipping deadline, where this level of front-loaded rigor may be unnecessary overhead.
Detail
procedure:
- 1. Define goals/KPIs (cost, weight, resolution, etc.) early and change them as little as possible
- 2. Design the hardest, riskiest parts first (e.g., route the tightest cable pinch point before finalizing the rest)
- 3. Over-iterate on the parts customers touch/interact with most (trackpad, keyboard) vs. under-iterate peripheral parts
- 4. Do anything you know you need to do right now — never bank time, because surprises always consume it
rationale: Each design iteration costs months; shipping days earlier than a competitor can be worth tens of millions; engineers rarely feel 'done,' so clear goals are what let you know you can ship.
In their words
Having your goals defined early and sticking to them is important. Hardware is not as adaptable to lots of changes throughout its development as anything digital.
Technology-arc transfer: VR as stepping stone to robotics
Source
Lenny's Podcast: Product | Career | Growth — Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple) (May 17, 2026) · at 2:44 ▸
When this applies
when a big technology investment fails to reach its original market but produces reusable underlying capabilities
The takeaway
Judge a technology investment not only by whether its original product category succeeded, but by whether it generated transferable foundational capabilities that unlock the next category.
Why it works
Core enabling technologies (spatial tracking, depth sensing, human perception modeling) developed for one product category transfer directly to adjacent categories once the direct market fails to materialize as hoped.
One level up
Apparent failures in a product category can be real successes in technology-capability building, visible only in hindsight.
Where it stops holding
This stops working when the invested technology is genuinely category-specific and has no adjacent application, so a failed product category really is just a loss.
Detail
approach: Reframe a product's commercial underperformance by asking what core technical capabilities it forced the team to build, and where else those capabilities are now load-bearing.
signals:
- heavy R&D spend with disappointing adoption
- technology stack (sensing, tracking, perception) has broader applicability
- new adjacent markets (robotics, AVs) emerging that need the same stack
rationale: VR didn't achieve mass adoption but Meta's SLAM, depth-sensing, and spatial-perception work is now central to how robots understand and move through space.
In their words
What I see now is in robotics, all of these technologies are being used because you need to understand how the robot is moving through space.