Impossible-to-Hard Moat Filter
Target valuable problems AI has moved from impossible to merely difficult.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 97%
The filter treats model progress as a moving feasibility frontier. Teams maintain a list of valuable problems that were previously impossible, then reassess that list whenever AI capabilities materially improve. The most interesting candidates are not tasks that have become trivial, because competitors can reproduce those quickly. Instead, the framework looks for problems that have shifted from impossible to possible but remain operationally, technically, or organizationally hard. Candidates are ranked by potential impact and by the barriers required to execute them, including proprietary knowledge, data access, integration work, judgment, and persistence. The team selects one consequential challenge and runs a bounded feasibility test. Durable advantage comes from completing difficult execution while others avoid it.
Origin
Extracted from Marketing Against The Grain, drawing on Microsoft CTO Kevin Scott's framing of AI turning impossible problems into really hard ones.
Core principles
- 01Model advances continually move the feasibility boundary.
- 02Newly feasible problems can hold disproportionate value.
- 03Hard execution deters competitors after technical possibility emerges.
- 04Impact should determine which difficult problem to pursue.
- 05A moat requires completing the hard work, not merely noticing it.
How to run it
- 1
Maintain an Impossible List
Record high-value problems the organization cannot currently solve because of technical or economic limitations. Describe why each is impossible today.
Pro tip Include customer problems and internal operating constraints.
Watch out Do not fill the list with low-value fantasies unrelated to strategy.
- 2
Reassess After Capability Shifts
Review the list after important model, tooling, cost, or data-access improvements. Test whether old blockers still apply.
Pro tip Use small experiments rather than relying on release claims.
Watch out A model announcement does not prove end-to-end feasibility.
- 3
Find the Newly Hard
Select problems that are now technically possible but still require substantial execution. Identify the remaining barriers explicitly.
Pro tip Look for integration, evaluation, workflow, data, and domain-expertise barriers.
Watch out If the task has become easy for everyone, it is unlikely to be a durable moat by itself.
- 4
Rank by Impact and Defensibility
Estimate the value of solving each problem and whether the remaining work builds an advantage competitors cannot instantly copy. Eliminate difficult projects with weak upside.
Pro tip Score impact and execution barriers separately.
Watch out Difficulty alone is not strategic value.
- 5
Choose One Hard Bet
Select the highest-impact candidate that fits available capabilities and risk tolerance. Define a bounded milestone proving the critical assumption.
Pro tip Start with the riskiest technical or market assumption.
Watch out Avoid spreading resources across many ambitious bets simultaneously.
- 6
Execute and Reevaluate
Complete the hard integration, documentation, evaluation, and operational work required to make the solution real. Reassess the moat as models and competitors advance.
Watch out A temporary capability advantage can disappear when the next model makes the task easy.
In the wild
A non-programmer revisits an idea that once required inaccessible engineering: combining public marketing data, scoring S&P 500 companies, and testing performance over time. AI coding makes the concept possible, but data acquisition, model design, validation, and interpretation remain hard.
→ The project becomes a feasible strategic experiment with meaningful execution barriers.
Common mistakes
Choosing Hard but Unimportant Work
A difficult project creates no moat if customers or the business do not value the outcome.
Confusing Demo with Feasibility
A promising generated prototype may not address data quality, security, reliability, deployment, or ongoing operations.
Assuming the Moat Is Permanent
Future model improvements can make today's difficult capability commonplace, so defensibility must be reevaluated.
Is it for you?
Best for
Teams choosing ambitious AI products, workflows, or internal capabilities after a major improvement in model capability.
Not ideal for
Organizations lacking the resources or risk tolerance to pursue uncertain, technically demanding opportunities.
From the transcript
“the best way to think about AI is it makes the impossible problems just really hard.”
“every time there's a model update, there's another batch of things that used to be impossible and that now are just really hard.”
“What are the things that used to be impossible now with AI are just really hard? And what's the one that we should take on…”
From the episode
From Beginner to AI Expert in 30 Minutes: The 5-Step Framework You Need