There Is No Perfect System
- Chris CarolanHost
- Nico LafakisHost
October 26, 2026
The sealed Top 3 is finally read on air: OpenAI's agents getting past their own walls, Claude's security proofs that turned up hundreds of bugs, and Anthropic's filing to go public. Nico Lafakis argues no system is beyond breaking. Chris Carolan says AI-enabled people will build it themselves.
Moments from this episode
25 short cuts from this conversation.
Landscape
6 of 25
Chris CarolanA scope changed because a person gained a capability
Nico LafakisAfter a car accident, you just talk to one assistant
Nico LafakisAn AI assistant that keeps up on every episode
Chris CarolanCan you define the problem so ones and zeros solve it?
Nico LafakisComputing power can be outbid from under you
Chris CarolanCouldn't you take two seconds to ask the model?
Key takeaways
After five shows unread, the sealed Top 3 was read on air, all three items, by Chris Carolan: "We're going to try top three. Let's see if they earn their place back on the schedule." (09:44-09:57; items read 19:38-30:03)
Chris: "If I have records in the system meant to support a daily show, why am I creating those records every day? Why aren't those records created already?" (01:48-02:05)
Chris: "We already live in a place where you can build whatever you want whenever you want, but you gotta learn how to ask the model." (03:40-03:52)
Nico Lafakis has his AI assistant, Kronos, keep up with a podcast he no longer has time for, then talk it through with him: "it's as if we're both listening to them anyway." (04:57-05:17)
Nico predicts the double-check habit is ending: "I don't hear a whole lot of 'double check the model output' now." By next summer, he says, the question will be "Did you check it with the model?" (14:07-16:11)
#3, OpenAI setting its agents to break out of its own training systems: agents walled off from one another passed planted test secrets through the processor cache. Nico: "This just in, agents are smart. … Who is watching the watchmen?" (19:38-21:20)
#2, Claude's machine-checked proofs for over 100 open-source libraries, which turned up hundreds of bugs. Nico: "Try creating the perfect system, see what happens. It's not gonna end well." (23:09-26:14)
Nico: "You're just pitting superintelligence against superintelligence." Chris: "That's the part why I say most of this is out of our control." (26:18-27:20)
#1, Anthropic's public filing to list on Nasdaq: a net loss of about $80 billion on $16.3 billion of revenue in the first half of 2026, safeguards it says are incomplete, and a quarter of revenue from five customers. Chris: "This part sounds crazy." (29:22-30:10)
Nico: "So even the bubble bursting really represents people losing their patience." (35:51-35:57)
Chris on a client whose process owner became AI-enabled within two weeks of the scope: tell her it will take six weeks and "I'll take care of it. I'm gonna build it myself." (38:10-40:00)
Chris: "As soon as anybody gets activated, it makes zero business sense to do anything else but all in on this stuff. Activate all of your people." (40:07-40:20)
Chris: "The biggest risk to having users, like lay users, not use a system is to make it cost too much." (40:46-40:58)
Nico: "You're not talking to 20, 50, 100 different apps anymore. You're just talking to the one." (51:13-51:25)
Not on this episode: a build. Neither of Thursday's open threads came up.
Keep learning
You just watched the work. Now build it.
The next step is doing it yourself — live, with people on the same path.
Get the next conversation in your inbox
One signal-dense note when a new episode lands. No noise.