Realized the initialization node could be removed so the system always starts directly in the medical debate, simplifying the architecture and reducin...Show more

Cameron shares how his graph kept reinitializing and overwriting results until he added an interrupt to break the loop in Langgraph Studio.

Cameron realized during demos they often skip discussing the code, prompting a shift to include deeper code walkthroughs in future presentations.

Forced the agent to engage in at least three turns with the patient after finding that a single interaction led to superficial diagnoses.

Encountered and debugged around ten issues including several catastrophic errors introduced by the initial automated configuration.

Tom discovered that the Claude CLI isn't supported on Windows when attempting installation during the podcast session.

Cameron teamed up with Flaud Code to rapidly prototype a LangGraph + Claude agent and demonstrate it live on the podcast.

Cameron’s key motivation was not just understanding the research but exploring how to turn it into a tangible, interactive AI product.

When no data was returned for ignored tests, the LLM swarm inferred failure and shifted hypotheses, prompting the team to create synthetic responses.

Microsoft’s applied medical AI paper shows a mixed-agent system outperforming both leading LLMs and human doctors on New England Journal of Medicine c...Show more

Tom ended up with a stack of Raspberry Pis that never got off the ground despite initial experimentation.

Cameron pranked friends by routing their calls through the Pixel’s AI answering machine to generate real-time AI-mediated responses.

Cameron quipped that Cloudflare’s ‘Are you under attack?’ prompt is possibly the best call-to-action he’s ever seen.

Figma filed its S1 in the past week and despite chatter about potential disruptors, survey data suggests it will maintain dominance in product design.

Tom invested in companies focused on storage and memory RAM anticipating high demand, but adoption was slower than expected, partly due to emerging al...Show more

Several AI-native and AI-enabled startups are driving significant sales growth purely through prompt refinement workflows and tailored evaluation metr...Show more

During development demos, intentionally displaying installation errors can signal transparency and reinforce that features are experimental.

Cameron accidentally triggered the jailbreak guardrail, which is meant to block bypass attempts, even though no bypass was attempted.

Cameron realizes a detail about their previous flight booking and considers making changes.

Cameron reveals he's used the AI tool more often than before, indicating growing adoption.