E62: How AI Agents Can Build While You Walk Away

In this live workshop session from the AI for Business event, team member Mike shows off a sales coaching and training platform he built entirely on his own, with zero coding experience, after seeing a similar tool Brian had put together. What started as curiosity turned into an all-night build session, roughly 150 prompts and sixteen planned phases, using Lovable to build the app and Perplexity Comet as an AI agent that could take his ideas, write the prompts, and execute the build in Lovable largely on its own. The app itself is built for sales teams and covers a lot of ground: a command center tracking calls and appointments, a role-play arena where reps practice against an AI voice trained to simulate different sales scenarios and then grades their performance, an intelligence dashboard tracking conversion trends and best calling times, an objection-handling library, leaderboards, a training academy for onboarding new hires, and a manager's coaching console. Brian and Mike also touch on the bigger lesson underneath the demo: treating AI less like a search engine and more like a collaborator you talk back and forth with until it surfaces ideas you wouldn't have thought of yourself.   Timeline Summary [0:01] When you actually need to bring in a developer, and why it's often much later than people think [1:29] Confirming you can build a full multi-user app without any developer help at all [2:24] Mike's origin story: seeing Brian's build and deciding to build his own version himself [3:35] The scope of the build: sixteen planned phases, uploading a company-specific knowledge base [4:27] Realizing the concept could extend far beyond one industry, into collections, customer service, and more [4:49] The all-night build session: roughly 150 prompts in one sitting [5:07] Introducing Perplexity Comet as the AI agent handling the build almost entirely on its own [6:13] Why Brian held back on sharing this approach at first, given the $200 a month cost of the tool [6:33] The progression from prompting AI yourself to letting an AI agent handle the whole process [7:14] Mike's own track record building apps in Lovable, including sports betting projects [7:34] The bigger lesson: talk to AI like a person and keep digging past its first answer [8:24] Touring the app: the war room command center and call history tracking [8:44] The role-play arena: text-based for now, with voice role-play already tested separately [9:33] How the AI-voice role-play grades a rep's pitch and gives improvement feedback [10:17] The intelligence section: outcomes, best calling times, and conversion trends [10:55] The objection bot, covering common objections reps can practice against [11:19] The leaderboard and personal performance report for individual reps [11:43] The manager's coaching console, including trend-based coaching assignments [12:05] The training academy, built for onboarding new hires with uploaded product training [12:38] The coaching console showing rep activity and live call listen-in capability [13:01] Wrapping the tour with the reporting section [13:41] What's coming in phase 13: real-time AI listening for objection handling during live calls   5 Key Takeaways You Don't Need a Developer Until You're Scaling — For most single-build apps, you likely won't need to bring in a developer until you're serving somewhere in the range of a thousand users. Before that point, your own time is usually better spent elsewhere anyway once the app is working. Multi-User Apps Are Fully Buildable Without Code — A complete app with individual logins, private databases, and end-to-end functionality can be built without any developer involvement, using tools like Lovable. AI Agents Can Now Build Almost Autonomously — The next stage after prompting AI yourself is letting an AI agent take your idea, write its own prompts, and execute the build in another tool like Lovable with minimal hands-on involvement from you. Talk to AI Like You're Talking to a Person — The single biggest lesson from this demo is treating AI as a real back-and-forth conversation partner. Don't settle for the first answer; keep digging to surface ideas you wouldn't have found otherwise. One Internal Tool Can Apply Far Beyond Its Original Use Case — What started as a sales coaching tool for one specific industry was quickly recognized as applicable to collections, customer service, and virtually any team that handles live conversations with customers.   Links & Resources Lovable (the AI app-building platform used to build the demoed app) — https://lovable.dev Perplexity Comet (the AI browser/agent used to autonomously drive the Lovable build) — https://www.perplexity.ai/comet ChatGPT (referenced as the model powering the voice role-play feature) — https://chatgpt.com Claude (referenced as a tool Mike used elsewhere, including for a fantasy baseball draft) — https://claude.ai   Enjoyed This Episode? If this demo made you realize you could build the internal tool you've been putting off, don't wait for the "perfect" build plan, start talking to AI about the problem and keep pushing past its first answer. Share this episode with a sales leader who's still doing rep coaching manually, and if it helped, hit subscribe and pass it along to a friend or colleague.