Loading summary
A
For years, I blamed the wrong thing whenever something didn't work. I told myself it was the idea or the timing or the market. It was none of those. Every business I run broke in the same exact place. Me. I build an AI crew to carry the business while I did. So that's what today is. I'm opening the garage door. This is my AI crew. Welcome to the Think AI Podcast. Each week we talk about the most exciting AI research tools, case studies, and more. I'm your host, Dev Goyer, and I've been working behind the scene in data and AI for over 30 years. Whether you are an AI expert, skeptic, or something in between, this podcast is for you. In my last episode of Grow Without Sacrifice, I told you a story about a hand bike that sat in my garage for six months while I worked myself into the ground. And Saturday, I finally rode it again. Because I build a AI crew, which is the set of AI employees to carry the business while I did. A lot of you sent me the same message over and over again, same version, which is, okay, Dev, this is a nice story. Now show me the actual thing. So that's what today is. I'm opening the garage door. This is my AI crew. One promise before we start. I'm going to show you what this AI crew does and why I build the way I did. I'm not going to hand you my secret recipe, the exact wiring inside each employee. But by the end, you are going to know exactly what's possible. And you will have the shape of it clear enough to start your own. Let's go. For years, I blamed the wrong thing. Whenever something didn't work, I told myself it was the idea or the timing or the market. It was none of those. Every business I run broke in the same exact place. Me. Here's what actually looks like. Every lead came to me. Every follow up waited on me. Every question, every scheduling clash, every decision, all of it landed on one desk. And that feels like control, but it's not. It's really a ceiling. The business could only ever move as fast as I could. And it stopped the second I did. When I slept, it slept. When I got sick, it got sick. So what I did that everyone tells you to do, I hire help. And I spend my days training them, queuing their work, checking their work. I turned one job into three. I hired agencies, got beautiful decks and not a lot of movement. I bought software and spend more nights configuring tools than running the company. And every one of those fixes had the same Flaw. They all still run through me. I just build a more expensive version of the same bottleneck. And that's when it finally clicked. I didn't need more hands to do the work. I needed something that I could actually own. I didn't need more people. I needed a system that could hold a role. Let me say the real problem in one sentence, because naming it is what unlocked the solution. The business could run a single function without me, not run worse, without me, could not run. There was no marketing function. There was just me doing marketing. There was no sales process. That was just me chasing deals. So I stopped asking the question everyone asked, which is what task I can automate. And I started asking a different question. What job could I actually fill? And here's the definition that changed everything for me. An AI employee is not a chatbot. A chatbot waits for you to type. An AI employee is four things. A job, a memory, a routine, and a folder it owns. It wakes up on its own schedule, it does the work, it writes back what it learned, and it briefs me in the morning. And that's the difference. It does everything. Automation does a task. Once an AI employee owns the outcome, quick gut check on who's listening because I want you to know I'm talking to you. If you are AI curious and you have played with ChatGPT and you are wondering if any of this is real for a business your size, it is. Stay with me. If you are an AI enthusiast, you have already built a few automations and you want to see what a real production setup looks like, this is one. I run my company on it. And if you are a real builder who's already thinking, I'm going to copy this, good. But I'm going to show you the shape so you can build your own, not clone mine. Yours should feed your business, not mine. So here's how it actually works. Day to day. I don't open an app, I declare an intent and the right part of the company wakes up. I say I'm doing build or marketing or sales or ops or finance and the system loads up that department and the right leaders step forward. Let me introduce the crew now because I want you to meet them like you would meet a real team. Maya is the chief of staff. She runs my morning standup and she triages my inbox and calendar before I even awake. Morgan, she's my cmo. She owns the content calendar, the publishing and the marketing strategy. The video you are watching went through her pipeline. Sasha, she is my VP of Sales Pipelines, proposals, sales strategy, CRM, HubSpot. Everything goes through her. The strategy, the execution and the monitoring. Owen, my VP of operation, manages delivery health vendors, keeping projects on track and deals with the firefighting situations. Felix, my cfo. Invoicing expenses, cash flow, Runway. What if the month end close? Everything goes through Felix. He's really a cfo, managing the business and keeping it healthy. Harper, my VP of hr. The roster, onboarding, tracking, who can do what in their performances. Theo, my cto. Theo's job is the crew itself. He maintains the skills, the rules, the plumbing that keeps everything running and everyone running. And behind all of them there's a built in for when I needed a brand new capability. One research is, one architect, one builds and one tester. Like a little product team that hires the next employee. Now here's the part I want you to be really clear about. Because people get scared by the word AI employee. Every one of these AI seats has a human counterpart in my team. Same task list, same definition of done. This doesn't replace my people, it multiplies them. Not bots in a drawer. A company with a front door. Now some of you will argue this is going to increase my work. No, it's not. They are doing the actual work. But there has to be someone managing when they go wrong or if something goes wrong. And that could be just one person taking all these AI employees reporting to them. It could be you. That's how I'm running the business for one of my businesses. Okay, so let's do a little bit of technical fork in the road. I'm going to keep this plain. To make AI do something useful, it has to talk to your tools, your systems, your email, your calendar, your CRM. And there are two ways to do that today. An API which is like calling a restaurant's kitchen directly and reading your order over the phone. It works great. But you have to know what specific kitchen rules are. Mcp, AI friendly and it's newer. Think of it as a standard doorway. So any tool can talk into kitchen the same way. One common language. So why pick one over the other? MCP is faster to wire up and it speaks one language across everything. The raw API gives you more control. And honestly, it often costs less when you are running a lot of volume. Our rule is simple. Use MCP when it's clean and already built for us. Drop down to the raw API when the MCP doesn't exist or it's flaky, or it's getting expensive. And I'll be honest about the trade off because Nobody tells you this part. You only see shiny objects on the road. MCP can hide your cost, and it can break the day a vendor changes something. The API is more work up front, but then you own it. Use the doorway when it's there, Build your own door when it isn't. Okay, so let's dig into the fun part. Here's the real example of that. I have a financial cockpit. My cfo, Felix, needs the book. But accounting system doesn't always hand a nice clean doorway. So here's what Felix does. He pulls the books, reconciles them against our CRM, and says what's actually closed and flex. Anything that doesn't match, and critically, anything that touches money still come to me for approval. But I don't want to leave it there because pulling a snapshot isn't the interesting part. The interesting part is what Felix built on top of these books. He doesn't hand me a spreadsheet. He hands me a cocktail. Let me show you, because this is the part I'm proudest of. And one note before I do, every number you see on the screen is blurred for the video. That's on purpose and that's my real company's money. But you don't need the numbers. You need the shape that tells you the whole story and help you build your own. The way Felix thinks is simple. The whole dashboard exists to answer three questions I ask all the time. Am I healthy? Which is my business is healthy? Can I reinvest? And what's coming in up in the header or the main page? He keeps two lenses side by side. One is accrual, which is structural profit. Is the business itself actually healthy. And the other one is cash, which is the survive and reinvest view. What literally goes in the bank. Same company, but two honest angles. Now let me walk you through the six views, because each one answers a different piece. First, the overview. This is the cockpit. Up top, there's a plain English verdict line. Something like healthy. This much reinvestable above the floor. And cash is still climbing. This is a language I like. You can change it. And under it, five gauges. Cash in hand revenue, ebitda, net cash flow, and reinvestable surplus are the Runway that could survive your company. The word floor matters. The floor is the cash reserve. Felix will never let me count as a spendable. It's the line he protects is the rainy day savings. The goal of this whole screen is one thing. Can I sleep tonight? And it answered it to me in one glance. Tab two, Profit and loss. Now, each of the books or online software has its own version of profit and loss. But in my world, this is different. This is a waterfall. It starts at revenue, steps down, then show the cost to deliver the work. Then overhead what we pay to the owners, the tax and then lands on the profit at the bottom. What I love is that Felix separates the profit the business is making and what we pay ourselves. And it is important and big because those are two different questions. This one tab answer Is the engine actually profitable before I take a dime out? That's it. Third tab cash flow. And here's a hard lesson every owner learns through its own painful way. Profit is not cash. You can be profitable on paper, still not be able to make your own payroll. So this tab shows cash in versus cash out month by month. Where the cash actually went. Broken into vendors, owner draws overhead, tax and how fast our clients are actually paying us. The goal is to catch the gap whether we are profitable or if we are out of cash before it catches me. 4th tab forecast if the others are the dashboard. This is really the windshield. It projects our ending cash in six months out, draw a downside line at break even and marks the floor down again. And the part that makes it real. It's life. Felix can change an assumption. Say revenue drop 20% or growth 50%. The whole line redraws in front of you. The goal is to see the valve while it's still far away. Not after I hit it. Fifth step, Arkansas and ap. In simple language, account receivable and account payable is who owes me and who I owe. The receivables are aged into buckets. Current 30, 60, 90 plus. And the 90 plus bucket is the one that bites. And it bites hard. That's money I earned. That's quietly turning into a problem. On the other side, the bills coming due. The goal is simple. Pull the money in before the money goes out. Sixth tab. This is about the owners. You. I call this the honest tab. It lays out what the partners actually draw. The split between salary and profit share, what we can keep in the business and how it is taxed. Because the real CFO question isn't can we pay to the owners? It's can we pay the owners without starving the business? And that's the real question. And this tabs keeps me honest about that. And then a bonus tab. There's a tab where I can just ask. I type a question in plain simple English. Now this is the really AI first approach. How much can I reinvest right now or how much I can save? For a larger project that is coming in future. Felix answers in one sentence and and with the actual numbers behind it. No spreadsheets plunking? I ask. He answers. So pull back for a second. The lesson from the integration still holds. And write this one down. You will almost never get the perfect integration. You get a good enough bridge and you put a human checkpoint on anything that moves money. But the bridge was never the prize. The prize is what Felix built on top of it. Felix doesn't hand me numbers. He hands me a decision. Now my favorite example because it's the one people light over others. I live in Microsoft 365 email, calendar and all of it. And we tried the Microsoft 365 MCP, the ready made doorway for what we needed. It had real limitations. So I had two choices. Wait around and hope it got better or build the piece I needed myself. So we wired directly into Microsoft's own Graph API for email and calendar can be extended to teams, chat and OneNote. And here's the payoff. Every job my crew runs on a schedule shows up on an actual event on a real calendar in my outlook. So I can open an outlook in the morning and literally see my AI employees working their shifts are on the calendar. And you will see it. The lesson A missing integration is not a stop sign. If the platform has any door at all, even a hard one, you can build the part you need. My calendar doesn't show you my meetings, it shows my employees shifts. So here's a mistake I made early so you don't have to the crew needs memory, but so do the humans on my team. And if the AI knows something my team doesn't, I haven't built a system. I have built a silo with extra steps. So there's one central place my humans actually live in. And in our example, this is notion. It manages tasks, projects, single source of truth, my daily standup, everything that is happening living there and the discipline that makes it work in this one source of truth. No duplicates. If a task lives in notion, it does not live in three different places which will quietly go out of date. The fastest way to break trust in a system is is to have two answers to the same question. So here's a quick example. My crew runs a whole content pipeline, drafting, designing work in canva short form video pipeline publishing out of the platforms. Now I'm still doing a lot of work. I'm still generating ideas, fine tuning those ideas, polishing those ideas through AI, recording the videos like this, cleaning them up, making sure Everything looks cleaner, nicer, and creates a story. But the real work is executed by my AI crew. And where a platform gives us a clean door, we push everything through one publishing layer. So a single piece of content fans out to LinkedIn, Facebook, Insta, or YouTube. I don't post four times, I approve once, and the crew ships it everywhere. But some platforms still won't give you a clean door at all. So for those, the crew drives an actual browser using a tool called Playwright and posed like a human would, clicking through the screens. It's not the same lesson from earlier, but when there's no API, you automate the front door instead. The crew doesn't make the work, it ships the work. Now, let me talk about the hard part, because I'm not here to sell you a fairy tale. My crew runs on Claude. And when Claude releases an update, things can quietly break. Permission gets reset, scheduled jobs lose their footing. Sometimes even the underlying program gets swapped out from under you. The first time this bit me, half my morning routine just didn't fire, and I didn't know why. And here's how I solved it. I created one sanction upgrade path. One script runs on a schedule every Friday evening. It upgrades cleanly through the proper channel, confirm if it's correct, and signed version checks that every scheduled job still point to something real, and regrants the permission my OS needs to let the crew work. And I can do a dry run on Saturday so that Monday morning everything is clean and fine. And the lesson is way bigger than Claude. A real system needs a maintenance ritual. Not wives, not I'll deal with it. When it breaks, you don't just build the thing, you keep it healthy on a schedule like any other real operation. Hand in hand with that, there's a single health check I can run anytime. It validates that my data is fresh, reconciles the book, and confirms all the pieces that systems are complete and consistent with each other. It's read only, totally safe, and I can run it whenever. And here's why that matters. The health check is the thing that lets me keep adding to the crew without the whole thing getting wobbly. Stable base first is the principle. Then I build the next route. On top of it, you earn the right to add the 10th employee by keeping the 1st 9th super healthy. Now, let's talk money, because this is where a lot of people's AI dreams die. Every word the AI reads or write cost you. The words are called tokens. Some words can be multiple tokens. And tokens are the meters running in the background. A bloated, sloppy system is an expensive system. It will quietly eat your budget. So here's what we did. We kept each employee's instructions short and focus on one job. We wired them to call each other instead of repeating the same information over and over again. And we ran a token saving layer on the heavy technical commands. So the routine stuff cost a fraction of what it used to. And the mindset shift is this. Treat your AI's attention like payroll. You don't drag every employee into every meeting. You only bring in who the job really actually needs. A lean crew isn't just cheaper, it is sharper. I want to be straight with you about the failures too. Things break. A vendor changes their API overnight. A scheduled job misfires. A model update subtly changes how an employee behaves. But the fixes is never heroic. Never. Me staying up till 2am is the same loop every time. Catch it in the health check, isolate the one that broke and then write the lesson down. The crew never repeats it once you fix it. The last part is the secret. My crew keeps a running log of its own lessons. The system gets smarter over time because it remembers its own mistakes. Not because I'm getting smarter, because the system is getting smarter. So now where is all this headed? A few honest fronts. First, a publishing layer. This is a real question of free versus paid tools. Figuring it out when paying actually earns versus keeping it on a subscription. You forget about. Second, MCP versus API. And I have done it already. So it's round two. As I grow, the game is lower cost and more control. And that actually means owning more of the connections. Myself and the big one I'm working on swapping some of the cloud calls for an open source models to cut correct cost on the work that doesn't need a top tier brain. Use the expensive brilliant model for judgment and advisory and writing. Use a cheaper model for the grunt work. Same as a real company. You don't put your most expensive employee on data entry. But here's the future that actually matters. It's not about me. What I just walk you through. Used to require a Fortune 500 budget, a whole back office, the whole operation team. A solo founder can now run that for a cost of a coffee a day. Now the door is open and it's not closing. The future isn't AI replacing the small business owner. It's the small business owner finally getting a team. I didn't build this crew to be impressive. I built it so so I could close the laptop at 5 o'. Clock. So I could be in the room when my kid walks in and wants to play a chess game with me instead of 3 mils deep so I could get back on that hand bike on a Saturday morning and feel the wind again in the sea. This is the crew. The crew is the How Grow Without Sacrifice is the why. You don't trade your health for the revenue. You don't trade your presence for growth. And here's the thing. Everything I just showed you today, I teach step by step on how to build your own AI crew inside Grow Without Sacrifice program. It's not automate your life hype. It's how to build a real team that runs your business without burning you to the ground. Go to growwithoutsacrifice.com and register. Because here's the truth I had to learn the hard way. You don't have to be the single point of failure anymore. And you don't have to figure it out alone. I'll see you in the next one. You have been listening to Think AI podcast with Dev. Take one idea from this episode and turn it into action.
Think AI Podcast | Ep. 11 – Meet My AI Crew
Host: Dev Goyal
Date: June 23, 2026
In this engaging episode of Think AI Podcast, Dev Goyal introduces listeners to his revolutionary concept of an "AI Crew" – a team of role-specific AI agents that collectively run his business operations. He shares actionable insights from his journey, discusses practical setups, technological choices, and pitfalls, and dispels common myths about AI automation. The goal: empower listeners—whether AI-curious, enthusiasts, or builders—to imagine and start creating their own AI-augmented organizations.
"Every business I run broke in the same exact place. Me." (01:00)
"The business could run a single function without me, not run worse, without me, could not run." (04:10)
“An AI employee is not a chatbot. A chatbot waits for you to type. An AI employee is four things: a job, a memory, a routine, and a folder it owns.” (05:18)
Dev introduces the AI crew as real team members:
Maya – Chief of Staff: runs stand-ups, triages email/calendar
Morgan – CMO: content calendar, marketing strategy
Sasha – VP Sales: pipelines, proposals, sales strategy
Owen – VP Operations: project delivery, vendor management
Felix – CFO: invoicing, cash flow, financial dashboard
Harper – VP HR: roster, onboarding, performance
Theo – CTO: maintains crew, skills plumbing
Plus: On-demand AI team for research, architecting, building, and testing new roles/capabilities.
Quote:
“Every one of these AI seats has a human counterpart in my team. Same task list, same definition of done. This doesn’t replace my people, it multiplies them… A company with a front door.” (11:20)
“Use the doorway when it’s there. Build your own door when it isn’t.” (15:00)
Timestamps: (16:40 – 24:00)
Natural language ‘Ask Felix’ tab: Plain English Q&A interface for financial decisions
Quote:
“Felix doesn’t hand me numbers. He hands me a decision.” (24:00)
Microsoft 365 example—when ready-made integration (MCP) didn’t work, the crew built a direct Graph API connection.
AI crew’s work appears as calendar events—“shifts” run autonomously.
“My calendar doesn’t show you my meetings; it shows my employees’ shifts.” (27:30)
Lesson: Lack of API is not a roadblock; tricks like Playwright allow browser-based automation.
Regular breakages from updates (e.g., Claude model updates).
Solution: Scheduled Friday upgrade scripts, health check verifications, dry runs over weekends.
“A real system needs a maintenance ritual... You don’t just build the thing; you keep it healthy on a schedule like any other real operation.” (34:20)
Continuous health checks ensure the crew scales reliably; add new agents only when base is stable.
Each “token” (word/action) costs money—sloppy design eats budget.
Solutions:
Mindset: “Treat your AI’s attention like payroll.”
“A lean crew isn’t just cheaper, it is sharper.” (38:00)
Breakages: APIs, scheduled jobs, model updates—all inevitable.
Fixes: Always check, isolate, learn, and document. The system’s running log builds “organizational memory.”
“The system gets smarter over time because it remembers its own mistakes. Not because I’m getting smarter, because the system is getting smarter.” (40:20)
Next steps: Smarter publishing, greater use of APIs for cost control, open-source models for non-creative work.
Key vision: Solo founders can now run a full back office—what used to require enterprise budgets.
“The future isn’t AI replacing the small business owner. It’s the small business owner finally getting a team.” (43:00)
“I built it so I could close the laptop at 5 o’clock... so I could be in the room when my kid walks in and wants to play a chess game...” (43:30)
Naming the Problem:
“The business could only ever move as fast as I could. And it stopped the second I did. When I slept, it slept. When I got sick, it got sick.” (02:40)
AI Employees vs Chatbots:
“An AI employee is not a chatbot... It wakes up on its own schedule, it does the work, it writes back what it learned, and it briefs me in the morning.” (05:25)
Why a Real Team, Not Bots:
“This doesn’t replace my people; it multiplies them. Not bots in a drawer. A company with a front door.” (11:20)
Hard Truths:
“A bloated, sloppy system is an expensive system. It will quietly eat your budget.” (36:20)
Dev’s presentation is honest, energetic, and motivational—mixing practical engineering lessons with philosophy. He’s both a guide and a peer, speaking from years of experience but also with humility about the continued challenges of making AI work in the real world.
Dev encourages listeners to stop being the bottleneck, consider the type of “jobs” in their business that an AI employee could own, and start building—not just for scale, but for freedom and quality of life.
End of summary.