1 min lesson
The AI-native paradox
Compare two rows from "The AI-native paradox", then say when each one fits.
Step 1 of 2
The AI-native paradoxIt cuts both ways in one loop
Cursor builds an AI coding tool and prizes people who instinctively apply AI to real work. Yet in their craft screens they famously bar AI assistance to test raw reasoning. Both are true at once and the same loop will test both.
- Where AI is barred
- Live technical/craft screen - defend your own reasoning on enrichment, routing or workflow design with no assistant.
- Where AI-native impact is the whole point
- Practical project - build an AI-agent automation against APIs or a Clay-style enrichment, scoring and routing flow that leans on LLMs.
- Where AI is barred
- Tests whether you actually understand the systems you claim to have built.
- Where AI-native impact is the whole point
- Tests whether you instinctively remove toil with agents and MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. instead of hand-running it.
| Where AI is barred | Where AI-native impact is the whole point |
|---|---|
| Live technical/craft screen - defend your own reasoning on enrichment, routing or workflow design with no assistant. | Practical project - build an AI-agent automation against APIs or a Clay-style enrichment, scoring and routing flow that leans on LLMs. |
| Tests whether you actually understand the systems you claim to have built. | Tests whether you instinctively remove toil with agents and MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. instead of hand-running it. |
Same person, two modes: raw reasoning when watched, AI-native impact when building.
Why the paradox exists
A company that sells impact cannot hire people who only know how to push buttons in a SaaS UI. So they verify the reasoning underneath first, then watch you apply impact on top of it. If you fake the reasoning, the practical round exposes you; if you can't wield AI, the screen looks fine but the project falls flat.