ExpertsFMThe most successful AI company you’ve never heard of | Qasar Younis
Lenny's Podcast: Product | Career | Growth · March 8, 2026
5 Selected Lessons:
framework · 2 heuristic · 2 reasoning_pattern · 1
Derive company values from what's already working, not philosophy
Source
Lenny's Podcast: Product | Career | Growth — The most successful AI company you’ve never heard of | Qasar Younis (March 8, 2026) · at 49:16 ▸ · at 1:12:53 ▸
When this applies
when any organization wants to define guiding values once it has early signs of what's working
The takeaway
Define organizational values inductively by identifying the concrete behaviors that are already producing success, rather than deductively from abstract ideals.
Why it works
Values distilled from observed causes of actual success are grounded and testable touchstones for hiring/promotion decisions, whereas abstract philosophical values have no empirical anchor and don't guide behavior consistently.
One level up
Normative frameworks (values, policies, culture) should be built from observed cause-and-effect, not imposed top-down from aspiration.
Where it stops holding
This stops working when the organization hasn't yet achieved any traction to observe causes from, or when early success was due to luck rather than repeatable behavior.
Detail
procedure:
- Wait until you have some early traction
- Write down the five to ten concrete reasons you believe you are succeeding
- Turn those reasons into named values
- Use the values explicitly in onboarding/new-hire meetings
- Assess, compensate, and promote managers against adherence to those values
benchmarks: {"example_values":["speed above everything / move fast move safe","never disappoint the customer","technical mastery","high output matters","laugh a lot","half the work is follow-up"]}
rationale: Values become living guiding principles, not abstract slogans, because they are literally the causes of the company's existing success and are tied to concrete, assessable behaviors (e.g., decisiveness under the speed value).
In their words
No, no, you should figure out why are we being successful? Like, literally, write down the five to ten things that are the reasons you are being successful and those become your values.
Balance open debate with decisive execution via explicit value pairing
Source
Lenny's Podcast: Product | Career | Growth — The most successful AI company you’ve never heard of | Qasar Younis (March 8, 2026) · at 1:04:46 ▸ · at 1:12:53 ▸
When this applies
when an organization needs both broad input-gathering and fast decisive action, which naturally conflict
The takeaway
When two necessary behaviors structurally conflict, don't pick one — encode both into a single evaluated value so the organization is forced to find the middle path case by case.
Why it works
Pairing two seemingly contradictory behaviors into a single named value (e.g., 'move fast, move safe') and then measuring/promoting managers specifically on both halves forces the organization to hold both in tension rather than collapsing into just one.
One level up
Sustainable culture change requires embedding tensions into measurable, compensated behaviors, not just stating them as aspirational values.
Where it stops holding
This stops working when the organization lacks the managerial discipline or measurement systems to actually evaluate adherence, making the paired value purely rhetorical.
Detail
procedure:
- Encourage everyone, especially junior/dissenting voices, to speak up with competing ideas
- Let the best idea win regardless of who proposed it
- At some point, information stops arriving — make the decision decisively
- Name the tension explicitly as a paired value (e.g., move fast, move safe)
- Assess and compensate managers specifically on both decisiveness and open input-gathering
benchmarks: {"named_value":"move fast, move safe"}
rationale: The middle way between total openness and overconfident decisiveness is hard to find organically, so it must be structurally engineered via values that are actually measured.
In their words
The decision in our values, that first value, the speed value, which is specifically the wording is move fast, move safe. That's specifically the wording. We assess our managers on adherence to those values.
Jobs 'lost' to automation are often jobs people already don't want
Source
Lenny's Podcast: Product | Career | Growth — The most successful AI company you’ve never heard of | Qasar Younis (March 8, 2026) · at 19:38 ▸ · at 26:54 ▸
When this applies
when evaluating fear of automation displacing labor in dangerous or undesirable jobs
The takeaway
Before treating automation of a job category as net job destruction, check whether humans are actually willing/available to do that job at the wages/conditions offered — unfilled, undesirable roles are not the same as displaced willing workers.
Why it works
Labor shortages in dangerous, undesirable jobs (trucking, mining, farming) already exist because workers now have alternative, more flexible options (e.g., gig work); automation fills a gap that isn't being voluntarily filled by humans anyway, rather than displacing willing workers.
One level up
Evaluate claimed social harms of a technology against the actual observed behavior of the people supposedly harmed, not against an assumed baseline.
Where it stops holding
This stops working when the automated job category genuinely does have willing workers who will lose income/purpose, not merely unfilled undesirable roles.
Detail
approach: Check revealed preference (are people actually taking these jobs?) before assuming automation destroys willing employment.
signals:
- average farmer age in late 50s signals upcoming labor shortage
- workers shifting from long-haul trucking to flexible gig work like Uber/DoorDash
- companies like McDonald's or mining firms reporting inability to hire
rationale: The trade-off of dangerous/inflexible work is no longer 'worth it' for workers who now have better alternatives, so automation fills a real gap rather than displacing willing labor.
In their words
If you look at farmers, the average age of a farmer is in their late 50s, 58 or so. What does that mean in 10 years from now? That means many of those farmers are going to be retiring if they're not already retired.
Read old books for signal, not new books for noise
Source
Lenny's Podcast: Product | Career | Growth — The most successful AI company you’ve never heard of | Qasar Younis (March 8, 2026) · at 58:02 ▸ · at 1:00:37 ▸
When this applies
when deciding how to allocate limited lifetime reading/learning budget across a flood of available content
The takeaway
When your consumption budget is small relative to the available content, prioritize content that has already survived a long selection/filtering process over unfiltered recent content.
Why it works
Time has already filtered low-quality content out of the pool of old books, so the average quality of a randomly chosen old, well-regarded book is far higher than a randomly chosen new book; given a lifetime reading budget of only ~50-100 books, maximizing average quality per book matters enormously.
One level up
Scarcity of attention justifies seeking pre-filtered signal sources over raw, unfiltered streams.
Where it stops holding
This stops working when the domain is changing so fast that old material is simply obsolete (e.g., current AI techniques), where Younis himself admits reading new material like the 'Vibe Coding' book 'goes against my grain.'
Detail
rule: Read old books, not new ones, because time has filtered out the noise; specifically seek out the best book in a domain you know nothing about.
threshold: Lifetime reading capacity estimated at ~50-100 books for the average person — too small to waste on low-quality/unfiltered content
rationale: Understanding of society and history, even indirectly connected to your industry, improves your capacity for judgment/taste as a leader, by Younis's own (unproven but strongly held) belief.
In their words
And in your life, you're going to, a thousand books, maybe you'll read. Like in the best case scenario, you're going to read probably 50 to 100 books, which is kind of crazy for the average person. So you're just not going read many. So don't read low quality content.
Bang-for-buck determines which physical AI use cases arrive first
Source
Lenny's Podcast: Product | Career | Growth — The most successful AI company you’ve never heard of | Qasar Younis (March 8, 2026) · at 19:38 ▸
When this applies
when forecasting which applications of a new general-purpose technology will be adopted first across many possible use cases
The takeaway
The first adopted applications of a new general capability will be those that can be layered onto already-mature infrastructure/hardware, rather than those requiring built-from-scratch systems (like humanoid robots).
Why it works
Use cases where most of the required engineering (the hardware/machine itself) already exists and only needs a layer of intelligence added capture nearly all the value of the underlying decades of engineering investment at low incremental cost, making them the highest-ROI early targets.
One level up
New general-purpose technologies diffuse fastest where they can ride on top of pre-existing sunk infrastructure investment.
Where it stops holding
This stops working when the emotionally compelling but infrastructure-light case (e.g. humanoids) benefits from novel demand/market dynamics that outweigh the infrastructure advantage.
Detail
rule: Expect earliest adoption where intelligence is added to existing, already-engineered machines (cars, mining rigs) rather than in novel form factors like humanoids, because the heavy engineering lift is already sunk.
threshold: n/a
rationale: Our brains love the humanoid concept emotionally ('because we're monkeys'), but pragmatically, leveraging 50-60 years of existing machine engineering is a far higher-leverage path for near-term intelligence deployment.
In their words
the reason is all that, you know, let's say engineering required to make this giant, you know, machine that moves dirt has already been done.