ExpertsFMScott Wu, Cognition
David Senra · June 28, 2026
5 Selected Lessons:
reasoning_pattern · 3 heuristic · 1 case · 1
Competitive drive as decision-tree thinking
Source
David Senra — Scott Wu, Cognition (June 28, 2026) · at 0:00 ▸
When this applies
when any high-stakes pursuit requires evaluating many branching future paths
The takeaway
Intense, lifelong competitiveness can be channeled into systematic exploration of options and consequences, turning emotional drive into analytical rigor.
Why it works
Treating a venture like a game tree search forces explicit consideration of move-countermove sequences rather than static planning.
One level up
Deep personal drives (fear of losing, need to win) can be redirected into productive cognitive habits if recognized early.
Where it stops holding
This stops working when the competitive drive becomes about ego/status rather than solving the actual problem, leading to overconfidence or burnout.
Detail
approach: View every decision as a branch in a tree search; calculate moves ahead, as in competitive games.
signals:
- hatred of losing
- obsessive calculation of outcomes
- treating company-building like a strategy game
rationale: Habits formed in childhood competition transfer into business strategic thinking.
In their words
I think strategy game I don't know. Even building a company, it feels the same. It's just like you're calculating the moves.
First-principles reasoning vs. historical pattern matching
Source
David Senra — Scott Wu, Cognition (June 28, 2026) · at 20:39 ▸ · at 22:56 ▸
When this applies
when forecasting any fast-moving technology or market where most people extrapolate from the recent past
The takeaway
In domains undergoing exponential or regime-shifting change, replace extrapolation-from-history with direct causal reasoning about what the mechanism allows, and ask why current limits wouldn't keep extending.
Why it works
Pattern matching on historical base rates works for stable regimes, but fails in the rare regime-change periods where the underlying drivers (e.g., exponential scaling) change qualitatively; first-principles reasoning asks 'why not further' rather than assuming limits.
One level up
Correctly identifying which 1% of situations are genuinely different from history (rather than defaulting to the 99% case) is where outsized insight and edge comes from.
Where it stops holding
This stops working when the underlying exponential driver stalls or saturates, making historical pattern-matching the more accurate guide again.
Detail
approach: Ask 'why can't this trend (e.g., AI autonomous task duration) keep doubling/extending' instead of assuming historical limits hold.
signals:
- a metric that has been doubling on a fixed cadence (e.g., every couple months)
- domain experts assuming current limits are permanent
rationale: The METR report showed AI autonomous-work duration doubling every few months, moving from seconds to hours — a trend worth extrapolating rather than dismissing.
In their words
the way that folks typically kind of predict the future or think about what happens next is they pattern match based on what they've seen historically
Focus as the startup's only real weapon against giants
Source
David Senra — Scott Wu, Cognition (June 28, 2026) · at 52:16 ▸ · at 54:48 ▸
When this applies
when a small player competes against a resource-rich incumbent in any market
The takeaway
Win against better-resourced competitors not by matching their resources but by concentrating entirely on one specific problem you will care about more than they ever can, because your survival depends on it and theirs doesn't.
Why it works
A giant spreading attention/resources across everything cares proportionally less about any single niche than a focused startup whose survival depends entirely on that niche, so the startup's intensity of care and speed of iteration compensates for the resource gap.
One level up
Asymmetric caring/urgency is itself a competitive resource, often more decisive than capital or headcount.
Where it stops holding
This stops working when the incumbent treats the niche as similarly existential (i.e., also 'bets the company' on it), eliminating the caring-asymmetry; or when the niche turns out not to be large/defensible enough to sustain a standalone business.
Detail
approach: Ask who has existential stakes in a specific niche vs. who treats it as one line item among many; bet on the side with existential stakes.
signals:
- incumbent has the resources but the niche is not core to their survival
- repeated historical pattern of 'why would X exist when giant Y already does this'
- startup's whole existence depends on winning this specific niche
rationale: Spotify beat Apple Music despite Apple's resources because Spotify would 'die' without success while Apple treated music as non-critical; the same logic applied to Datadog/Snowflake/Databricks vs. the clouds, and to Cognition vs. Microsoft.
In their words
the way that you build a real kind of like you know, real, like, lasting business or a lasting product is by really, really focusing and narrowing on one specific thing
Humans are badly calibrated for exponential change
Source
David Senra — Scott Wu, Cognition (June 28, 2026) · at 17:55 ▸
When this applies
when evaluating how much an exponential trend will change a market or society that people instinctively underestimate
The takeaway
Expect people — including experts — to underestimate how much and how fast an exponential technology will reshape normal life, because their evolved intuition is linear.
Why it works
Human risk/reward intuition evolved for roughly linear, bounded payoffs (e.g., a hunt feeding you for days), not for scenarios where a single advance could be equivalent to 'a thousand years of food,' so we systematically underestimate compounding change.
One level up
Calibrate your forecasts explicitly against the fact that human intuition defaults to linear thinking, and consciously correct for it.
Where it stops holding
This stops working when the trend in question is not actually exponential/compounding, in which case linear intuition is appropriate.
Detail
rule: Assume the public (and even insiders) will underestimate the speed and scale of exponential technological change.
threshold: A trend doubling every few months compounds to orders-of-magnitude change within just a few years — treat this as the default expectation, not an outlier.
rationale: People got used to transformative appliances and lifestyle changes quickly but originally found them jarring; the same adaptation pattern will likely repeat with AI.
In their words
People often underestimate, I think, how fast things can change and how fast the world could change.
Releasing an imperfect prototype (Devin launch at 13% SWE-bench)
Source
David Senra — Scott Wu, Cognition (June 28, 2026) · at 30:18 ▸ · at 34:16 ▸
When this applies
when deciding whether to ship/announce a new category-defining product before it is fully reliable
The takeaway
When you believe you're early to a real paradigm shift, demonstrating genuine (even flawed) capability publicly can be more valuable than polishing privately, as long as you don't overpromise.
Why it works
Publicly planting a flag on a bold claim about the future — even with an imperfect product — builds brand, recruiting pull, and customer curiosity faster than waiting for perfection, because being first to articulate a new paradigm has compounding reputational value.
One level up
Being structurally late to a market can be overcome by being first to commit visibly and correctly to where the market is going, not by matching incumbents' resources.
Where it stops holding
This stops working when the gap between demo and reliable product is so large that it destroys trust, or when the 'hot take' about the future turns out to be wrong.
Detail
situation: Cognition was founded over a year after ChatGPT and years after OpenAI/DeepMind/Anthropic existed, with no resources, brand, or customers — 'no right to win.'
decision: They publicly demoed Devin's real (if cherry-picked, better-than-average) runs, including the MongoDB setup success, rather than waiting until it was polished, and did not retreat to a safer chatbot-style product despite heavy public criticism.
outcome: Devin went viral, 13% SWE-bench beat prior state of the art by 3-4x, and despite 'no customers, no revenue' at launch, the company grew from roughly $1M to $500M in revenue in about 18 months by being first to plant the 'AI as coworker' flag.
lesson: Startups win not by matching incumbent resources but by committing early and visibly to a correct read of the future and executing hard on it.
In their words
the reason that you are sometimes able to anyway is if you really, like, you know, plant your flag in the ground and and and put a stake into, like, what you think the future is, and you run like hell towards that