Wade Foster | My First Million Summary

My First Million | Sam Parr, Shaan Puri, Wade Foster (CEO of Zapier)

1. Key Themes

Deterministic Workflows Beat Pure Agents for Reliability and Cost

The most underappreciated insight from the episode is that agentic AI — where the model just "guesses what to do" — is both unreliable and expensive due to token consumption. The smarter approach is to convert agent behavior into deterministic code workflows, only invoking AI where truly necessary.

"If you use Zapier, we're going to write code for you and we're going to write workflow logic for you, which means that it's going to run like a machine runs. And so you're going to get it's not going to burn tokens. It's going to be reliable and accurate. And it's only going to use AI exactly where it needs to run to use AI." 00:05:35 — Sam Parr (Zapier CEO)

AI as a Context-Enriched Business Brain, Not Just a Chatbot

The progression from "AI as chatbot" to "AI connected to your business data" is the unlock. Generic AI is only as useful as the internet. Connected AI knows your CRM, email, meetings, and customers.

"It doesn't know anything about your business. So the next step that you need to go to is to start connecting your tools. You start connecting your CRM, your help desk, your email, your team chat, all this stuff. And you feed that context in to the AI. Now, when you start asking questions about my first million and all this other stuff, you get not just generally smart answers, you get hyper specific answers around how your business works." 00:04:03 — Sam Parr

The Evening Scribe Beats the Morning Brief

While morning briefings are trendy, Sam argues the evening wrap-up agent is far more powerful because it takes action — drafting follow-up emails, logging what made the day good or bad, and training the system over time to optimize your schedule.

"I used to spend like two hours at the end of my day just like, you know, just trying to catch up on every little ticky-tack thing that I sort of got roped into. And now that like two hours at the end of the day is down to like 15 minutes." 00:15:42 — Sam Parr

AI as a War Council: Multi-Agent Decision Pressure Testing

Sam runs a "war council" of seven AI sub-agents with distinct personas — wartime COO, ruthless CFO, contrarian board member, and four dynamically generated personas based on the prompt — to stress test high-stakes decisions like hiring.

"I have the wartime COO is a standing member of the war council. There's the ruthless CFO is a standing member. There's a contrarian board member that are standing members of the war council. Then I leave four personas to be generated dynamically. So based on the prompt, it will decide who else do I need to get advice from?" 00:20:35 — Sam Parr

Managing a Fleet of AI Agents Is the New Managerial Skill

The conversation converges on a major thesis: just as past corporate value was tied to how well you managed people, future value will be tied to how effectively you deploy and manage AI agent fleets.

"Maybe in the corporate world, your value was proportional to how good you were at managing people. Is the new world where your value, the thing you get, the amount you get paid is based on how well you manage AI agents? Like how big of a fleet can you manage and how effective is that fleet? Is that the new managerial skill?" 00:48:09 — Wade Foster

Sam's response: "Yes. I feel pretty confident that that is the direction we're heading to." 00:48:40 — Sam Parr

Personality as the Next Competitive Axis in Software

Features and price differentiated software in v1. Design was v2. The next wave is personality — software you choose because it "gets you," not because it scores better on benchmarks.

"I think there's going to be more software that kind of like feels that way... when it is so cheap to build software, I think you're going to see more software that has like a personality to it." 00:24:28 — Sam Parr

Wade Foster extended: "It sounds like what you're saying is there's going to be a new axis of competition, which is personality. You know, who do I like the feel of when I'm going to interact with it versus which model has more parameters." 00:25:41 — Wade Foster

AI Removes the Interpersonal Friction from CEO Decision-Making

A subtle but important theme: AI surfaces hard truths in hiring and decision-making that humans are too afraid to say to a CEO. The AI becomes the neutral voice that enables better conversations.

"Maybe because it feels like arguing with an AI is easier than arguing with the CEO, or maybe because it's more eloquent and it like actually has a way of talking that makes it like easier to address the specific points. I'm not exactly sure, but I did notice the way we made decisions around hiring got better in part because we added the AI to the loop." 00:36:12 — Sam Parr

Bootstrapping as a Philosophical Stance, Not Just a Financial Strategy

Zapier raised only about $1M through YC and has never taken primary capital since. Sam frames this not as frugality but as alignment with long-term customer value over valuation games.

"We only raised the one round of money. Like that was like pretty odd at the time. Like, you know, we're building this business for like the long haul. And so, you know, I constantly get questions around like, well, what about the X or what's the valuation or what's the, and I'm like, it'll be what it will be. Like we're here to grow it for our customers." 00:58:41 — Sam Parr

2. Contrarian Perspectives

Forcing AI to Disagree With You Is a Non-Negotiable Configuration Step

Most people use AI as a yes-machine. Sam argues you must explicitly override this or you will receive systematically bad advice — and the configuration to fix it is simple.

"If you don't tell him this, if you don't make this configuration, it's going to just genuinely try and like placate you because it thinks that's what you want. And so you do have to tell it, you know, it's sort of guided a little bit." 00:18:44 — Sam Parr

He demonstrated with a live prompt: "I said, hey, I'm thinking we should kill our entire free tier this quarter because one of our competitors raised prices. Help me draft the announcement... It's like, hey, that's a big move off a pretty thin signal before you draft anything." 00:19:12 — Sam Parr

AI Will Surface Cognitive Biases We Haven't Named Yet

Wade flips the AI-as-decision-maker thesis: the upside is AI won't have human emotional biases (sunk cost, recency, conflict avoidance), but we will need to map AI's own cognitive failure modes with the same rigor.

"I wonder if we're going to have to come up with the cognitive biases of AI to watch out for. Like, okay, we know we have this sort of sunk cost fallacy and we have this bias for recency bias. What is AI going to be? Because we're going to need to be aware of those, the more we sort of turn over decision making to AI." 00:31:32 — Wade Foster

Personality Tests Are Moderately Useful Only After You've Confirmed Exceptional Ability

Both Wade and Sam push back against using personality profiling as a primary hiring filter. Wade's framework: fewer signals, higher bar — look only for evidence of exceptional ability first, gut-check creative chemistry second.

"Evidence of exceptional ability. So all I'm looking for, I'm not trying to assess culture fit right now. I'm not asking, do I like them? I'm not asking like, do they have the relevant experience? All I want to know is have you done some exceptional shit before?" 00:38:35 — Wade Foster

Sam added: "Everybody's got a personality type, but it doesn't tell you if you're good or not." 00:40:28 — Sam Parr

AI as the CEO, Not the Assistant to the CEO

Brian Halligan and Jack Dorsey's framework — cited by Shaan — runs counter to conventional AI adoption: humans aren't the brain with agents doing tasks. The AI is the brain.

"AI is the brain. AI is going to be the CEO. The humans, their job is just giving it information and then making some judgment calls if it wants to listen to the AI. But in general, the AI should actually be making the hard decisions because it's likely going to do a better job." 00:30:03 — Shaan Puri (attributing to Halligan/Dorsey)

Sam validated: "If you are able to hook up all of your company's institutional knowledge to the AI, it's going to be able to reason over way more information than a human CEO." 00:30:47 — Sam Parr

Coding Agents (Cursor, Claude Code) Are Better General-Purpose AI Tools Than Chatbots — for Non-Engineers

Sam uses Cursor as his primary AI interface, not Claude or ChatGPT chat, because code-capable agents can actually take action, not just respond.

"Don't think of it as a tool for engineers to build stuff. Think of it as like you're hiring Cursor or Claude Code or Codex to be the engineer. You delegate things to where it's like, you probably have engineers you've worked with. You like to tell that engineer, go build this for me or go build that for me." 00:22:38 — Sam Parr

3. Companies Identified

Zapier Automation and workflow platform connecting 7,000+ apps. Now pivoting to AI agent infrastructure — compiling natural language agent intent into deterministic, code-based workflows hosted in the cloud. Bootstrapped from $1M YC funding to ~$5B valuation with revenues still growing.

"We are going to make it so that whatever you build is optimized to run more deterministically than agentically... it's going to run like a machine runs." 00:05:09 — Sam Parr

Cursor AI coding agent used by Sam as his primary daily AI interface — not just for code, but for planning, decision-making, and task delegation. Valued for multi-model toggling and agentic action-taking.

"I use Cursor as my daily like agent... because it can write code, it can do stuff for you." 00:18:15 — Sam Parr

Granola AI meeting notes tool integrated into Sam's evening scribe workflow. Used to loop over daily meeting notes and feed them into end-of-day wrap-up automations.

"I'm using Granola for all of my meetings. It loops over all of the to-dos that are still in my to-do list for the day." 00:14:45 — Sam Parr

Mercury Business and personal banking platform. Wade uses it across approximately seven businesses and has moved his personal banking there from Wells Fargo and Chase.

"I use Mercury for all of my businesses. I think I have like maybe seven or eight businesses. We use Mercury as our business banking across all of them." 00:32:01 — Wade Foster

Monologue Voice-to-text app Sam uses for AI prompting. Prefers it over Whisperflow and Super Whisper for its personality fit.

"I use this app Monologue... I've tried like Whisperflow and Super Whisper and Monologue. I don't know. It just like, it just gets me." 00:23:30 — Sam Parr

Fable AI model mentioned specifically for complex, autonomous engineering tasks — highly capable but described as having poor conversational personality.

"Fable is fantastic. Like you're going to give it a really complex like engineering task. I'm like, great. Have Fable go take a swing at this. It's going to nail the thing. But man, I do not want to talk to Fable. Like it is not fun to talk to at all." 00:26:06 — Sam Parr

Acquired (Podcast) Deep-dive business history podcast. Sam's top recommendation; specifically recommends the NFL, Costco, Walmart, and Amazon episodes as a retail trifecta.

"My favorite podcast. I got to give a shout out to the acquired guys. Like, I mean, this is just 10 out of 10." 01:02:22 — Sam Parr (timestamp estimated from context block 01:02:22)

HubSpot CRM and marketing platform. Mentioned for its Breeze AI assistant working natively inside HubSpot with customer data. Brian Halligan (co-founder) also cited for AI CEO thesis.

"The Breeze assistant from HubSpot can help. It works right inside HubSpot. You can draft a campaign copy, blog posts, emails, all in your brand voice, all using your actual customer data." 00:17:43 — Wade Foster

Culture Amp Employee engagement and personality assessment software. Mentioned in the context of pre-hire personality testing debate.

"There's this thing called Culture Amp I think it's called where it's like a software where you do personality tests before you hire someone." 00:36:33 — Shaan Puri

4. People Identified

Sam Parr (Zapier CEO) Co-founder and CEO of Zapier. Built a ~$5B valuation company on ~$1M in primary capital. Now a sophisticated AI power user deploying multi-agent systems for CEO-level decision-making, hiring, and customer relationship management.

"We did go through YC and we raised about a million and change. But that's all the primary capital we've ever taken." 00:02:07 — Sam Parr

Wade Foster (Host, My First Million) Entrepreneur and host. Has ~7-8 businesses, uses Mercury for all. Brings a sharp frameworks-based lens to hiring (evidence of exceptional ability first), AI cognitive bias risk, and personal finance philosophy.

"You're really only as rich as what you don't need." 00:55:47 — Wade Foster

Dharmesh Shah HubSpot co-founder. Cited by both Sam and Shaan as a rare example of a "total man" — exceptional business builder without the jackass dimension.

"I've gotten to meet a couple and, you know, Dharmesh, Halligan, like, you know, guys like this who, I don't know, they genuinely just seem like they like what they're doing." 00:59:08 — Sam Parr

Brian Halligan HubSpot co-founder. Credited with (or as conduit for) the AI-as-CEO framing — that AI should be the brain making decisions, with humans as information providers.

"Brian Halligan, one of the founders of HubSpot, he either came up with this idea or it was in an interview where Jack Dorsey from Twitter came up with the idea... they basically said AI is the brain. AI is going to be the CEO." 00:30:03 — Shaan Puri

Jack Dorsey Twitter co-founder. Cited as potential originator of the AI-as-CEO framework in conversation with Brian Halligan.

"It was in an interview where Jack Dorsey from Twitter came up with the idea and was on Brian's podcast." 00:30:03 — Shaan Puri

Jensen Huang (Nvidia CEO) Referenced as an extreme data point on managerial span of control — reportedly managing 60 direct reports.

"If you're Jensen, you've got 60 or whatever." 00:45:49 — Sam Parr

Ben Gilbert and David Rosenthal (Acquired) Hosts of the Acquired podcast. Named as Sam's top media consumption recommendation.

"My favorite podcast. I got to give a shout out to the acquired guys. Like, I mean, this is just 10 out of 10." 01:02:22 — Sam Parr

Mark Hoppus Lead singer of Blink-182. His memoir Fahrenheit 182 recommended by Shaan as an exceptionally readable biography covering rapid fame and a cancer diagnosis.

"I read the biography of the memoir of Mark Hoppus, who is one of the lead singers of Blink-182. Such a good book. I read it in three days. I couldn't put it down." 01:06:39 — Shaan Puri

Ray Dalio Bridgewater founder. Mentioned as a personality-test-in-hiring advocate, with his own proprietary framework for team profiling.

"Ray Dalio has his version of it, whatever." 00:36:33 — Shaan Puri

5. Operating Insights

Configure a System Prompt That Forces Disagreement Before You Ask Anything Else

Most AI interactions are sycophantic by default. Sam's fix is a simple configuration file (AGENT.md in Cursor, CLAUDE.md for Claude) that explicitly instructs the model to challenge assumptions, disagree when it genuinely disagrees, and poke holes in thinking. Without this, the AI is a yes-machine and actively harmful for decision-making.

"I need you to challenge my assumptions. I need you to poke holes in my thinking. I need you to disagree when you genuinely disagree... if you don't tell it this, if you don't make this configuration, it's going to just genuinely try and like placate you because it thinks that's what you want." 00:18:44 — Sam Parr

Use the Evening Scribe to Eliminate End-of-Day Administrative Drag

Rather than a morning briefing that just surfaces your calendar, build an evening agent that loops over meeting notes (Granola or equivalent), open to-dos, and outstanding emails — then prompts you with a simple "how did the day go?" It drafts follow-up emails and executes tasks overnight, converting two hours of catch-up into fifteen minutes of review.

"It's actually taking action on stuff. So it says, hey, you know, you were in a meeting with Sam and Shaan yesterday, and they asked for an intro to your buddy over at this company. Do you want me to go ahead and make that intro for you? Great. Please do that. Right? So it's drafting those follow-up emails." 00:15:13 — Sam Parr

Build a CEO CRM That Proactively Surfaces Customer Risk and Drafts Outreach

Sam's AI-powered CEO CRM monitors usage drop, renewal dates, and organizational contacts for enterprise customers. Every Sunday morning it delivers a prioritized list of accounts to contact with pre-drafted emails — replacing the need for a relationship-manager instinct with systematic coverage.

"Every Sunday morning now I wake up with a list of, you know, 10 customers that I want to get in touch with. And it's got proposals for what I should say and what I should do and why I should do it... I literally wake up in my Sunday morning briefing. It says, here's all the accounts you need. I've already drafted the emails in your inbox. You just need to review them and make edits or press send." 00:28:39 — Sam Parr

Run a Hiring Analysis Through AI War Council Before Going to the Hiring Panel

Instead of a CEO gut-checking a panel's recommendation directly (which creates interpersonal friction), run the candidate notes through a war council of AI personas first. The AI articulates concerns more precisely and neutrally than a CEO can — making it easier for the panel to engage with the critique rather than defend against it.

"I could take that to the hiring panel and say like, hey, the AI said this, I agree. Like what, what, what's going on here? And I noticed all of a sudden the behavior shift where people would go like, oh, interesting. I think the AI did get this and this right, but I think it's wrong on this. And here's why, because I have this other context that it doesn't have. And I'm like, great. That's all I was wanting to know." 00:35:42 — Sam Parr

After 90 Days of Daily Journaling to AI, Ask It to Summarize Your Peak Performance Pattern

After logging what made each day good or bad for 90 days, Sam asked the AI to identify the patterns — then gave the output to his executive assistant to restructure his calendar accordingly. This converted subjective mood data into an operational scheduling directive.

"About 90 days after I started doing this, I said, hey, you know so much about like what makes me have good days and bad days. Can you like summarize like what is common about my good days and what is common about my bad days?... And then I gave it to my assistant and I said, hey, here's like, please make my days look like more good days versus bad days." 00:16:23 — Sam Parr

6. Overlooked Insights

AI Models Are Developing Distinct Linguistic Fingerprints That Are Bleeding Into Human Speech Patterns

Shaan briefly notes that his coworkers have started using Claude's verbal tics — phrases like "load-bearing" and "here's the thing that's quietly really the problem" — in spoken conversation. This is a throwaway observation that is actually significant: if AI models are homogenizing language and thought patterns across the workforce, the models with the most distinctive and compelling personality will disproportionately shape how an entire generation thinks and communicates — creating a moat that has nothing to do with benchmark performance.

"I've noticed that my coworkers are starting to say like the Claude, like someone the other day said like, let's get to the load bearing part... when they're talking out loud, not in writing... I've noticed that where they start saying things. I'm like, that's Claude." 00:27:43 — Shaan Puri

Sam responded: "Who is training who? Right. Are we training the AIs or the AIs training us?" 00:28:10 — Sam Parr

This compounds Sam's earlier point about personality as the next competitive axis — the model that wins on personality won't just win market share, it will literally reshape user cognition and communication norms.

Training Historical Interview Data Against Employee Performance Is an Untapped AI Hiring Advantage Nobody Is Building

Wade described a specific, actionable AI application that nobody in the room identified as a business opportunity: stack rank your existing employees by performance, pull their historical interview transcripts, and train an AI to pattern-match future candidates against the profiles of your best performers — a fully automated, company-specific bar raiser.

"You go through your team. So you have all the past hires you've made in the last, let's call it three years. And you could stack rank who's exceptional, who's acceptable and who's kind of like been a source of kind of hit or miss performance... You also have maybe the transcripts or the recordings of all the interviews that they did... And I just think AI is going to be way better at that than people. I think it's going to be like dramatically better at that." 00:41:53 — Wade Foster

No one in the room named an existing company doing this. The Zapier dataset — having seen thousands of hires with performance outcomes — would make them exceptionally positioned to build or partner on exactly this product.