Knowledge-Work End-State Idea Generation
Pick an area of knowledge work and reason forward to its end state as AI matures — then build for that.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 82%
Before Cursor, Truell's team ran a structured ideation exercise: rather than building models, they picked areas of knowledge work and asked what each would look like as AI matured over a decade. For each area you ask what the end state of the work is, how the tools change, and how the models need to improve to support it. This forward-reasoning-from-the-end-state method is what surfaced software creation as the opportunity.
Origin
Michael Truell describing Anysphere's founding idea-generation exercise (late 2021–early 2022).
Core principles
- 01Don't just build models — pick an area of knowledge work and think about how AI transforms it.
- 02Reason from the mature end state backward to what to build now.
- 03Scaling laws mean AI will improve even with no new ideas, so bet on capability trajectory, not just today's tech.
- 04Ambition is a differentiator: a crowded space can still be wide open if incumbents aren't reasoning far enough ahead.
How to run it
- 1
Enumerate areas of knowledge work
List domains of knowledge work and treat each as a candidate to be transformed by maturing AI.
- 2
Imagine each area's end state
For each, ask what the end state of the work looks like, how the tools to do it change, and how the models must improve to support those changes.
Pro tip Anchor on scaling-law evidence that models keep improving just by scaling data and compute, even absent new ideas.
- 3
Judge the ceiling and the ambition gap
Prefer areas with a high ceiling and lots left to build, and look for spaces where existing players aren't being sufficiently ambitious about where things are heading.
Pro tip A leapfrog is possible when the ceiling is high and incumbents under-reach — 'too late' can still be wide open.
Watch out Truell's team first mis-chose a 'sleepy, boring, uncompetitive' area (mechanical engineering) to avoid competition and had to reverse — avoiding competition is not the same as picking the right end state.
- 4
Commit to the area you'd dedicate your life to
Choose the area where you're genuinely excited and see the biggest transformation, then build the tool for its end state.
Pro tip Being the end user of that area sharpens every subsequent decision.
In the wild
The team ran the exercise, first picked mechanical engineering because it seemed uncompetitive and sleepy, spent four months, then 'came to our senses' — they saw programming players weren't being sufficiently ambitious about how AI would blow through software creation, and switched to building Cursor.
→ They landed on one of the fastest-growing products ever ($0 to $100M ARR in ~20 months, $300M ARR by year two).
Common mistakes
Optimizing for low competition over the right end state
Picking a 'sleepy, boring, uncompetitive' area (mechanical engineering) to dodge competition led to four wasted months in a field they weren't excited about and didn't understand.
Dismissing a hot space as 'too late'
Programming already had GitHub Copilot, but incumbents weren't ambitious enough about the end state — treating a crowded space as closed misses leapfrog opportunities where the ceiling is high.
Is it for you?
Best for
Founders choosing which domain to build an AI product for during a capability-driven technology shift.
Not ideal for
Builders optimizing a known product in a mature, low-ceiling market where the end state is already reached.
From the transcript
“how are each these areas of knowledge work uh going to change in the future as this tech gets more mature. Like what is the…”
“even if we had no new ideas, AI was going to get better and better just by pulling on simple levers like scaling up the…”
“we decided to work on you know uh an area uh of knowledge work that we thought would be relatively uncompetitive and sleepy and and…”
“they weren't being sufficiently ambitious about um where everything was going to go in the future and how kind of all of software creation was…”
From the episode
The rise of Cursor: The $300M ARR AI tool that engineers can’t stop using
Michael Truell (co-founder and CEO)