As AI agents prepare to reshape commerce, four technical primitives may promise to fix the Internet’s original sin: an economy built on unverifiable attention.
Warburg is trying to solve capitalism’s problem (alignment of incentives, verification of value) using capitalism’s tools (markets, cryptographic proof, optimization).
But the core problem isn’t technical. It’s that extractive systems can never be made just through better verification. You can verify that a laborer worked 8 hours, but if they’re paid $2/hour, verification doesn’t make it just.
Similarly: you can verify that an AI agent fulfilled an intention, but if the intention system crystallizes vulnerable desires, if the Context Graph is controlled by a stranger, if the attribution algorithm is opaque—then verifiable ≠ just.
The covenant move: Stop trying to make extraction just. Build systems where extraction is structurally impossible.
This requires:
∙ Desires that circulate within relationships, not broadcast to strangers
∙ Memory held by people who have mutual obligation to you, not by platforms
∙ Contribution negotiated through dialogue, not calculated through algorithms
∙ Futures designed by democratic deliberation, not optimized through simulation
These are slower. They’re less efficient. They’re also the only way to protect human dignity at scale
Your core insight—that attention markets need verification—is correct. But verification alone isn’t the solution. You need structural alternatives.
The Intention Graph could work if it were:
∙ Local (not global)
∙ Relational (not anonymous)
∙ Covenantal (not contractual)
The Context Graph could work if it were:
∙ Controlled by the person and their trusted community
∙ Subject to right-to-be-forgotten
∙ Auditable for accuracy
The Attribution system could work if it were:
∙ Designed democratically (not by engineers)
∙ Contestable through repair processes
∙ Weighted toward flourishing (not just efficiency)
The Simulation could work if it were:
∙ Explicitly political (not pretending neutrality)
∙ Subject to continuous validation against actual human experience
I don't think most intents are explicit in either user behavior or linguistic interaction (chat or search), but are implicit and contextual. That doesn't make them unavailable - but LLMs would have to engage in reverse prompting (and be good at it) to explore the deeper motives behind intents (always the grail of marketing).
Do you think LLMs could be capable of exploration such that a context of intents, implicit and perhaps even unarticulated by the user, could be captured in graph form?
And on context graph, do you think organizations might obtain social context graph information from AI use within the org such that they can build knowledge management insights and relationships from the ground up, that is, from use? (Who asks and shares what knowledge, search, topical exploration etc with whom, for what project, etc....)
Intent capture is structurally impossible for consumer purchases. It’s an output from, not an input to, the shopping journey. I don’t know what I’ll accept until I see it. This agentic approach works for B2B commodity procurement—find me 1000 10 mm M8 screws. It doesn’t work for the consumer market where most of your projected trillions live.
Warburg is trying to solve capitalism’s problem (alignment of incentives, verification of value) using capitalism’s tools (markets, cryptographic proof, optimization).
But the core problem isn’t technical. It’s that extractive systems can never be made just through better verification. You can verify that a laborer worked 8 hours, but if they’re paid $2/hour, verification doesn’t make it just.
Similarly: you can verify that an AI agent fulfilled an intention, but if the intention system crystallizes vulnerable desires, if the Context Graph is controlled by a stranger, if the attribution algorithm is opaque—then verifiable ≠ just.
The covenant move: Stop trying to make extraction just. Build systems where extraction is structurally impossible.
This requires:
∙ Desires that circulate within relationships, not broadcast to strangers
∙ Memory held by people who have mutual obligation to you, not by platforms
∙ Contribution negotiated through dialogue, not calculated through algorithms
∙ Futures designed by democratic deliberation, not optimized through simulation
These are slower. They’re less efficient. They’re also the only way to protect human dignity at scale
Your core insight—that attention markets need verification—is correct. But verification alone isn’t the solution. You need structural alternatives.
The Intention Graph could work if it were:
∙ Local (not global)
∙ Relational (not anonymous)
∙ Covenantal (not contractual)
The Context Graph could work if it were:
∙ Controlled by the person and their trusted community
∙ Subject to right-to-be-forgotten
∙ Auditable for accuracy
The Attribution system could work if it were:
∙ Designed democratically (not by engineers)
∙ Contestable through repair processes
∙ Weighted toward flourishing (not just efficiency)
The Simulation could work if it were:
∙ Explicitly political (not pretending neutrality)
∙ Subject to continuous validation against actual human experience
∙ Revisable when it produces harm
I don't think most intents are explicit in either user behavior or linguistic interaction (chat or search), but are implicit and contextual. That doesn't make them unavailable - but LLMs would have to engage in reverse prompting (and be good at it) to explore the deeper motives behind intents (always the grail of marketing).
Do you think LLMs could be capable of exploration such that a context of intents, implicit and perhaps even unarticulated by the user, could be captured in graph form?
And on context graph, do you think organizations might obtain social context graph information from AI use within the org such that they can build knowledge management insights and relationships from the ground up, that is, from use? (Who asks and shares what knowledge, search, topical exploration etc with whom, for what project, etc....)
Intent capture is structurally impossible for consumer purchases. It’s an output from, not an input to, the shopping journey. I don’t know what I’ll accept until I see it. This agentic approach works for B2B commodity procurement—find me 1000 10 mm M8 screws. It doesn’t work for the consumer market where most of your projected trillions live.