Extract the Chain-of-Thought, Not the Recommendation
Treat every advisor as an LLM: mine their reasoning, not their verdict, because their answer is trained on a different corpus.
- Difficulty
- Easy
- Time to result
- ~ongoing to results
- Steps
- 3
- Confidence
- 88%
A model for processing advice from smart people. Everyone is effectively their own LLM, trained on a different corpus of experience — so their literal recommendation is a one-size-fits-all output that may not transfer to your context. The high-value signal is their chain-of-thought: WHY they reached that conclusion. Inspect the reasoning the way you'd inspect a reasoning model's chain-of-thought, then re-apply that logic to your situation, which may yield a different but better-fitted answer.
Origin
Howie Liu's 'meta learning' from years of receiving scaling advice, framed through the reasoning-model analogy. He connects it to Brian Chesky's 'founder mode' (from Lenny's episode and the YC retreat), noting Chesky's specific decisions worked for Airbnb's context but the reasoning is what transfers.
Core principles
- 01Everyone has different priors — we're all our own LLMs trained on a different corpus (ServiceNow vs Facebook vs Airbnb experience).
- 02A recommendation is a one-size-fits-all verdict; the reasoning behind it is context-portable.
- 03Ignoring advice from smart people is wrong; blindly executing their literal recommendation is also wrong.
- 04The 'why' can yield a different outcome when applied to your context — and that's the point.
How to run it
- 1
Don't take the recommendation at face value
When a smart, experienced person tells you 'do X' (scale up this way, hire these operators, eliminate the PM role), resist both reflexes: neither dismiss it nor execute it literally.
Pro tip Remember the advisor isn't incompetent — they had real reasons — but those reasons came from a different training corpus than yours.
Watch out Blindly trusting 'here's the action you should take' from many people at once produces contradictory, context-blind decisions.
- 2
Interrogate the reasoning
Ask why they reached that conclusion — inspect their chain-of-thought the way you'd read a reasoning model's. Get the underlying causal logic, not just the verdict.
Pro tip The 'why did you do that?' is more informative than the decision itself; make it your default follow-up.
- 3
Re-apply the logic to your context
Take the extracted reasoning and reason forward from your own priors and situation. Accept that the correct answer for you may differ from theirs even when the logic is sound.
Pro tip Chesky eliminating PMs at Airbnb (replacing them with program managers and product marketers) made sense for Airbnb's context; the transferable part is the reasoning about product-led leadership, not the org-chart move.
Watch out Copying the surface decision because it worked at a famous company ignores that your corpus and constraints differ.
In the wild
Liu heard Brian Chesky's founder-mode principles and, rather than copying Chesky's specific choice to eliminate traditional PM roles at Airbnb, extracted the reasoning — a CEO must play a CPO role, care deeply about product, and make holistic non-incremental bets.
→ Liu independently deduced similar principles and applied the reasoning to Airtable's own context (the fast/slow reorg and CEO-as-IC posture), reaching decisions fitted to Airtable rather than importing Airbnb's org chart wholesale.
Common mistakes
Copying a famous company's decision verbatim
A decision optimized for another company's corpus and constraints can misfire in yours; without the reasoning you can't tell whether it applies.
Dismissing advice because it doesn't fit as stated
Rejecting the whole recommendation throws away the valuable reasoning inside it; the fix is to extract the 'why,' not to ignore smart people.
Is it for you?
Best for
Founders and leaders receiving high volumes of conflicting advice from investors, operators, and peers while scaling.
Not ideal for
Genuine domain experts giving hard-constraint guidance (legal, safety, compliance) where the literal instruction, not just the reasoning, must be followed.
From the transcript
“it's almost like we're all our own LMS right and like we all have different training from a different corpus of data”
“that like chain of thought like why did you recommend this is actually more informative than the actual like just do this recommendation”
“the why actually was very informative and then be able to take that and say like okay like how would I apply that and maybe…”
From the episode
How we restructured Airtable’s entire org for AI
Howie Liu (co-founder and CEO)