Source
Y Combinator Startup Podcast — Max Junestrand: You Need The Willingness To Learn Faster Than Anyone Else (August 29, 2026) ·
at 29:50 ▸
When this applies
when deciding whether to invest heavily in a custom, point-in-time solution versus riding the improvement curve of a general-purpose platform you don't control
The takeaway
When a core enabling technology is improving fast, don't spend scarce capital trying to engineer around its current weaknesses — spend it on building close customer relationships and distribution so you can capture value as the technology catches up.
Why it works
If the underlying general-purpose technology is improving exponentially, resources spent hard-coding around its current limitations become stranded cost once the next generation makes that workaround unnecessary; better to invest in distribution and delivering today's value while staying positioned to absorb tomorrow's capability for free.
One level up
Allocate resources based on which curve you're betting on, not on patching today's gap.
Where it stops holding
This stops working when the underlying technology's improvement trajectory stalls or when the specific gap you're ignoring never actually gets closed by the platform.
Detail
rule: Don't fine-tune/build proprietary workarounds for a fast-improving general model; build for 'the world today and maybe one step ahead' and focus on customer relationships and delivering value now.
threshold: Tribal knowledge in 2022-23 said fine-tuning was 'the name of the game'; Legora instead chose zero fine-tuning partly from necessity (no money) and partly from conviction models would keep improving.
rationale: I think we recognize that you don't have to build for the world over here, like you need to build for the world today and maybe like one step ahead.
In their words
I even remember, I think it was Bloomberg, who spent millions and millions and millions of dollars building a Bloomberg law model. Our view was partly because we didn't have enough money and partly because we truly believe that to be right, the models will keep improving.