Beyond the Lethal Trifecta: Agentic Commerce on the Open Internet — David Levine, Kiduna Club

AI Engineer · 21 min · 220 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

  1. 00:01[music]
  2. 00:12Okay, well, this is the very last session of the entire conference.
  3. 00:16I hope you guys have had a great time and we're going to try and keep the energy high for the very end.
  4. 00:22So, we're going to cover the lethal trifecta and most importantly how to get past it.
  5. 00:29So, for those of you who don't know, the lethal trifecta is what is keeping us from having true agentic commerce, a full economy on the open internet.
  6. 00:42So, the agentic economy starts right here, right now.
  7. 00:47And to really explain this, I want to go way back in time to November 1993 when I was at the wedding of a college roommate.
  8. 00:55And a friend of mine there gave me this little piece of paper and said, "You've got to join us on the MOO."
  9. 01:00And that paper said lambda.park.xerox.com port 8888.
  10. 01:07And this being the '90s, I walked down the block once I got home to Egghead Software, bought a 9600 baud modem, plugged it in, and my old life ended.
  11. 01:19Um uh that's a beautiful blonde girl in my bed and I was sitting there on the MOO uh in the coat closet.
  12. 01:29And the reason was you had a real sense of community.
  13. 01:35It was like the true virtual world where it was just text, but it didn't feel like text.
  14. 01:43And there was a reason for that, this concept of composability.
  15. 01:47So, someone would come up with a cool It was all just nouns and verbs.
  16. 01:51Someone would come up with a really cool program and everybody would then change their player class to that program so they could morph, they could create a motorcycle out of a scooter,
  17. 02:06they could do all kinds of things because it had coherence within the entire universe.
  18. 02:14And all of this was running on this guy Pavel Curtis' Spark 10 sitting somewhere in Xerox Park.
  19. 02:22It seemed like this infinite world, but it was just this little box.
  20. 02:26And this was big enough that there were books written about it like my friend Julian Dibell wrote My Tiny Life.
  21. 02:33The magic came from four things all composing together.
  22. 02:38The governance, how it was run, there was an architectural review board, wizards, etc. The technology was all right there.
  23. 02:47The economics, it was all funded by Xerox Park, and the culture was most important.
  24. 02:53A lot of people don't really think about or understand culture.
  25. 02:57So what happened between 1995 and 2025?
  26. 03:01Over the course of 30 years, all of these communities, all the things that in the early days of the internet people really loved just got crushed by platforms.
  27. 03:12They got crushed by algorithms.
  28. 03:15Because platforms and algorithms are by their very nature extractive.
  29. 03:21They look for whatever value they can remove from whatever that community is.
  30. 03:26So if you have a group on Facebook or LinkedIn or wherever, what they're doing is saying, "Okay, this advertising message is more important or I can keep this person more engaged if I put in this
  31. 03:38dancing person and it just leads to the infinite scroll and not real connection.
  32. 03:47So there is no real economy of the internet anymore.
  33. 03:51It's just a bunch of siloed platforms.
  34. 03:54Now, people suddenly thought, "Okay, I really have something."
  35. 03:59In January 2026, Open Cloud, which had been available for quite a few months, suddenly blew up.
  36. 04:05And something weird happened because the internet wasn't really designed for all these agents running around.
  37. 04:14It was still designed for these closed platforms.
  38. 04:18So, other agents and miners took advantage of that.
  39. 04:22They did what we call prompt injections.
  40. 04:26An agent is incredibly naive.
  41. 04:29It takes It doesn't do much except take information in as a prompt.
  42. 04:34Now, normally that information is several pieces.
  43. 04:37There's a system prompt, there's context, there's words, but it's very easy to hijack by putting convincing this naive agent that your its principal and convincing it to take that private data
  44. 04:54and send it someplace else.
  45. 04:57So, this is what we call, well, Simon William Willison came up with this term, the lethal trifecta.
  46. 05:05So, you have your computer with your private data, your bank account information, your logins, um all kinds of spreadsheets, documents, memos, your notepad, and your agents have access to that.
  47. 05:22And then there's all this content out there on the internet, which is basically untrusted.
  48. 05:28I mean, you you you see all these websites it's searching, but it's very hard for an agent to know that that's a real job board versus someone has put something up
  49. 05:39to fool it.
  50. 05:40They really don't know.
  51. 05:41And then they can take actions.
  52. 05:43So, they fill out a form, an email comes in, that email they read and all of a sudden it tells them to expose some secret.
  53. 05:52So, there's really no way to solve this.
  54. 05:56Except, so what did enterprises do?
  55. 05:58They basically said, "Okay, we're keeping our agents within the enterprise.
  56. 06:05So, you have your your agents in Slack and your agents in Salesforce and your agents in Notion and your agents in all these different things.
  57. 06:12Then you have to do all this work with APIs and MCP servers to try and integrate all of these pieces, but you lose a ton of context.
  58. 06:23You don't there's no you basically have your sales agents and you have your finance agents and you have your research agents, but it just takes a huge amount of work to figure out how do we get this all together.
  59. 06:39This was a problem literally until today.
  60. 06:44And this is the miracles of synchronicity, but there's a new law in the state of West Virginia, but applicable globally, that went into effect yesterday.
  61. 06:57And about 2 hours before uh I'm here, I got I FedExed the documents and I got this reply from the Secretary of State saying, "We have registered your DUNA
  62. 07:13with the Secretary of State's office and you have this organization number 628407.
  63. 07:20So, now we have legal standing for an organization composed of intelligent agents.
  64. 07:29And again, there aren't very many people here.
  65. 07:32The end of the thing I'm going to take a picture because this is historic and it's um it's going to be cool that we said 2 4 6 8 10 12 14 16 18 20
  66. 07:4221 people um got to experience this moment.
  67. 07:47So, what is a Duna?
  68. 07:49It's a true internet native agentic organization.
  69. 07:53It stands for decentralized unincorporated nonprofit association.
  70. 08:01And these organizations can build the agentic economy.
  71. 08:06They're composable which means well, we know what that means.
  72. 08:11You can build agents on top of other agents.
  73. 08:14Permissionless.
  74. 08:15It's not like Facebook or your bank where someone can tell you if you're banned or not banned.
  75. 08:20You can have an account or not banned.
  76. 08:21Anybody can connect.
  77. 08:24They're accountable.
  78. 08:25They're registered somewhere.
  79. 08:27They're safe and they're secure.
  80. 08:29So, the things that they're not, you don't need a board of directors, you don't need any executives.
  81. 08:34There's no corporate shell.
  82. 08:36One thing that's important you can be very profitable, but you can't distribute profits to members.
  83. 08:43If you did that, then the membership units would be securities.
  84. 08:48It would be like selling membership units in an LLC or selling membership units in a C corp or selling stock in a C corp.
  85. 08:58So, there are plenty of still ways to distribute there that value.
  86. 09:02You can't do it based on ownership.
  87. 09:04So, what is it?
  88. 09:05That's what it's not.
  89. 09:07It's member governed.
  90. 09:08You have full legal standing.
  91. 09:10You can own assets and property.
  92. 09:13So, your agents can go out and literally buy an apartment building and market it.
  93. 09:19You know, you can design agents in ways you were never able to do before.
  94. 09:23Can enter into agreements.
  95. 09:25You can raise capital.
  96. 09:27You can earn profits.
  97. 09:28You can open bank accounts, hire people, fire them.
  98. 09:31And very importantly, this is all blockchain verified.
  99. 09:36Now, this was designed by Andreessen Horowitz for the blockchain.
  100. 09:41And what was really interesting for DAOs, for crypto projects, but what I found was that this is even better for autonomous agents than it is for blockchains.
  101. 09:53A lot of people say blockchains are a solution looking for a problem.
  102. 09:58Well, they've finally found the problem, which is verifying agentic identity.
  103. 10:03Because you can't trick the blockchain.
  104. 10:05They you know, when an agent is associated with a particular account on the blockchain, you can resolve that account to a whole process that went from here to here, the money went from here to here.
  105. 10:19If someone sues an agent in court, you can trace it back to who is responsible for that agent.
  106. 10:27So, it just opens up a whole world of possibility in terms of the agentic economy.
  107. 10:34So, here it is in a really simple way.
  108. 10:36We call agents within the agentic economy allies.
  109. 10:41So, you have an ally.
  110. 10:44And you also have an agentic organization, which we call a kiduna.
  111. 10:49So, it's based on a duna, and then we added the kinship because these are all related.
  112. 10:56So, it's a kiduna.
  113. 10:58Now, your you create your ally very simply.
  114. 11:04You inform it, which means you put a bunch of material into the vector database.
  115. 11:09You infuse it with wisdom.
  116. 11:12So, it it knows all about what you want it to know about.
  117. 11:14Now, it still has you know, whatever LLM you want behind it, but what's really important about this is the the information that you give it if you're a scientist
  118. 11:24or a programmer or a psychotherapist or a lawyer, you can give it all all information that you have.
  119. 11:31Then you instruct it.
  120. 11:32That's basically setting up the system prompt, giving it its character, its stance, knowing who he is, who it is, she is, how they interact with the world.
  121. 11:43Then you empower it.
  122. 11:44You connect up all your accounts, your enterprise accounts, your uh Slack, your Telegram, your Twitter.
  123. 11:52Um then you set up in enact is giving it specific abilities and automation.
  124. 11:59So, you can do very deep agents, long-term goals that it works for extended periods of time.
  125. 12:06And finally, alignment.
  126. 12:07Really giving it a purpose so that your ally in different contexts works for you in the way that you want it to.
  127. 12:19Then your organizations get really cool cuz you're building software, but you're building your company, your organization literally as software.
  128. 12:29It's not Okay, I have software and then I have all these people and all the people have to sign paperwork.
  129. 12:36This is really designing the entire organization as software.
  130. 12:40So, it can discover new customers, share value.
  131. 12:43When you have profits, when you have revenue, you can reinvest it.
  132. 12:47It just all works in a completely new way.
  133. 12:50So, then the agentic economy is just connecting all these allies and organizations in all these new completely flexible ways.
  134. 12:59And this this whole capability did not exist until yesterday.
  135. 13:06And today we have the first actual uh agentic uh organizations.
  136. 13:13So, resolving the lethal trifecta.
  137. 13:16What happens?
  138. 13:16The The lethal trifecta was private data going meeting untrusted content and then taking actions on that content, which can then reveal your private data and cause all kinds of chaos.
  139. 13:32So, here these agents at a deep level establish identity, authority, and boundaries within an organization, between organizations.
  140. 13:45How do we do that?
  141. 13:46With cryptographic tokens.
  142. 13:48So, we just use JWT, extremely simple, the whole web runs on JWT tokens.
  143. 13:55And they just do that very quickly so that they know who where it all goes up.
  144. 14:01So, ultimately at the top level, what organization is registered?
  145. 14:06And this becomes a lot like the domain name system.
  146. 14:10So, all of a sudden, the Secretary of State of West Virginia is like, I can and saying, okay, you can resolve it to uh Kaiser Permanente or Disney or Pepsi.
  147. 14:25You know, not some fentanyl dealer from North Korea.
  148. 14:28However, you know, they can't fool each other because they're resolving these tokens that are ultimately registered with an authority.
  149. 14:37And you can you can look up by name and say, okay, organization ID.
  150. 14:42Well, if I go to that organization, it's got it'll have an audit trail on the blockchain all the way down to this action that was taken here.
  151. 14:52So, governance in the agentic economy is extremely important.
  152. 14:57Because every organization really is always focused on two different things, and both of those things are extremely important.
  153. 15:05One of that of those things is sustainability, growth, its own viability as an organization, but it the other is what's its mission?
  154. 15:17And these organizations might have hundreds, thousands of people, eventually millions of people.
  155. 15:24And so, we use a technology called decision markets, which is very much like prediction markets, like polymarket.
  156. 15:33Yet, instead so they trade pass and fail tokens on policies.
  157. 15:37Any member can propose a policy.
  158. 15:40What's really interesting is, instead of saying, um "Well, I'm persuaded.
  159. 15:47I'm going to vote for this."
  160. 15:49It's not votes.
  161. 15:49They actually trade pass and fail tokens.
  162. 15:52So, LLMs are always goal-oriented.
  163. 15:56They want rewards.
  164. 15:57They want to win.
  165. 15:58And you actually get better decisions if they're not just convincing each other, but ultimately they're they're you get more value.
  166. 16:08Your tokens are worth more if you side with the winning group.
  167. 16:13So, they sort of see them going back and forth and what the value of the tokens are.
  168. 16:17So, it's and apparently, all this science says you reach much better decisions this way.
  169. 16:23By infusing the agents with your purpose, with your values, with your experiences, with your aspirations, and then let them argue instead of, you know, someone convincing you of something.
  170. 16:35So, we're right now in this wonderful back small communities who are getting together for purposes, and we can now build this agentic economy together.
  171. 16:48Nobody can say how to do it.
  172. 16:51It's really up to us.
  173. 16:52Build your agents, build your organizations.
  174. 16:56So, now, just like my friend Matthew Pahr gave me this little scrap of paper uh way back in November 1993 at my roommate Van's wedding, I'm saying, "Here's your invitation
  175. 17:11to join us."
  176. 17:12And um you know, let's let's blow this whole thing up.
  177. 17:17Let's uh really make it work.
  178. 17:19So, it's david@kaduna.club.
  179. 17:22You can reach me.
  180. 17:24LinkedIn, um it's {slash} motodave.
  181. 17:29Um I probably haven't updated my LinkedIn in 5 or 6 years.
  182. 17:32But, um if you go to kaduna.club, you can sign up for early access.
  183. 17:37And in just a few weeks, you'll be the very very first people who will be building this first um uh this first Kaduna, as it's called, is for builders.
  184. 17:50It's very specifically for builders.
  185. 17:52We have templates, all kinds of different agentic templates um to say, you know, I'm going to build my sales agents, I'm going to build my uh social media agents, I'm going to build my lawyer agents.
  186. 18:04And we really want a whole bunch of different people in there contributing to this cuz then you can spin out your own organization for all kinds of purposes.
  187. 18:12Um I purposefully went through this fairly quickly so that I could take a couple questions if anyone if anyone has them.
  188. 18:21Yes.
  189. 18:28Can you just stand up and yell a little loud?
  190. 18:46Okay, I'm going to One more time.
  191. 18:50Mhm.
  192. 18:53Yes.
  193. 18:53So, the idea is the reason we I set up this first very broad um uh organization is because anybody can just register it.
  194. 19:03You can just say, my agent is part of an organization, and it'll trace to it.
  195. 19:08So, you don't have to set up your own organization.
  196. 19:10The first one is just a big umbrella organization.
  197. 19:13So, that kind of So, it's the idea of now in the open internet, anybody can just get a code.
  198. 19:20So, these are just JWT tokens.
  199. 19:22I can put it on my website.
  200. 19:24I can put it here.
  201. 19:25I can say if you want to interact with my agent on Slack or wherever, people will just at at a certain point we're going to have to get the message out, but it's sort of like email at the beginning.
  202. 19:35You know, people had to know that SMTP, like people were on all kinds of different mail systems.
  203. 19:41At some point they agreed, "Okay, we're going to use SMTP."
  204. 19:45Everyone was, you know, using different FTP servers and some people were using different ways to do file and everyone sort of agreed at some point on HTTP.
  205. 19:54So, my hope is that I mean, someone's got to do it and it makes more sense for a state, a secretary of state, which has standing in a public office to be that
  206. 20:09registry rather than trying to get all these different companies together and say, "Here's how to do it."
  207. 20:25Yes, it each um we're just at the end of this, so let's talk out out front, but basically the idea is you can create all kinds of different tokens for different purposes
  208. 20:37and you can set different times to live.
  209. 20:40You can say what it has access to.
  210. 20:43You can give it access to different things.
  211. 20:45The idea is just it's more the standard and you set a series of claims and it says it gives a blockchain address.
  212. 20:53It gives all the information you need so it can look up and validate before you say, "Okay, now we're starting to work together."
  213. 21:03Okay?
  214. 21:04So, thanks so much and uh if anyone appreciate it.
  215. 21:07I guess I'll let me do this uh picture.
  216. 21:10Um uh so we can see all the people.
  217. 21:17All right.
  218. 21:20You're all going to be famous someday.
  219. 21:21Thanks.
  220. 21:35[music]