My First Million · Episode Brief
I built 4 Agents in 55 minutes to save 20+ hours a week
Wade Foster built Zapier by making automation accessible to non-engineers, and the agents demo in this episode suggests we're in the same moment again — except the stakes are higher.
Wade Foster co-founded Zapier in 2011 by betting that non-technical people would build their own integrations if you made the interface simple enough. That bet was correct, and Zapier built an enormous business on it. Now, in 2025, Foster is watching the same bet play out one layer up: AI agents that can execute multi-step tasks across applications, built by people who have never written a line of code. The episode is structured as a live demo, which creates a different kind of accountability than a conversation.
The Instant Dossier demo — building an agent that compiles research on any person or company on demand — runs in under six minutes and produces something that would have required an hour of manual research. The value proposition is immediately legible even to a non-technical viewer. Sam builds it, narrates what he's doing, and doesn't hide the moments where something doesn't work the first time.
Model Context Protocol is the most technically dense section of the episode, and Foster explains it in a way that holds up. MCP is essentially a standardized way for AI models to call external tools — a universal connector layer that means you don't have to build custom integrations for every combination of model and service. The business implications of MCP becoming a standard are significant: whoever builds the best MCP connectors will own significant distribution leverage in the agent stack.
The inbox zero agent and the employee fraud detector demos extend the logic from the dossier: agents work best when the task has a clear input, a defined output, and a human who can verify the result. The getting-your-team-to-use-AI section is where Foster is most useful — the tactical reality of organizational AI adoption is almost always messier than the demo suggests, and Foster is honest about what actually works versus what the demos make look easy.
Key Ideas
- →Wade Foster's thesis is that we're in the same moment for AI agents that 2011 was for no-code automation — the tools are mature enough to be useful but not yet obvious enough that everyone is using them, which means early movers compound disproportionately.
- →Model Context Protocol represents a potential standardization layer for the agent ecosystem — a universal connector that, if it wins, will matter the way HTTP mattered for the web.
- →The Instant Dossier demo illustrated a category of use that Sam called 'the research tax' — the portion of every knowledge worker's day spent on information assembly that produces no new thinking.
- →Getting teams to actually use AI requires a different approach than getting individuals to use AI — Foster's argument is that adoption lives or dies on whether the first use case fits the workflow of the least technical person on the team.
- →The employee fraud detector demo is the most counterintuitive of the four — using an AI agent not to create efficiency but to create surveillance capacity that previously required dedicated compliance staff.
Worth Remembering
Sam built the Instant Dossier in real time and it produced a research summary that was noticeably better than what most human researchers would generate in the same time — the demo worked, which is not always guaranteed.
Foster described a moment when a Zapier customer reported that their agent had handled 2,000 customer service interactions before anyone realized no human had reviewed a single response — and that the responses were better than average.
The Model Context Protocol explanation used an analogy that clicked: 'it's like USB for AI tools' — the value is not in the connector itself but in everything that becomes possible when the connector is standard.
Wade Foster admitted that after 14 years building automation tools, he still finds the current AI agent moment surprising — and that the people who are most surprised by capability progress are usually the ones who've been watching it most closely.