Via Lenny’s Podcast

“For example, think about if your objective is to hike Half Dome in Yosemite, you don’t start hiking it every day; you start training yourself for minor parts and then you slowly improve [until] you go to the end goal, right? I feel like that’s extremely similar to build[ing] AI products.

You don’t start with agents with all the tools and all the context that you have in the company in day one and expect it to work. You need to deliberately start in places where there is minimal impact and more human control, so you have a good grip of the current capabilities and what [you can] do with them. And then slowly, you lean into more agency and lesser control.

So this gives you that confidence: I know that this is the particular problem that I’m facing, and the AI can solve this extent of it, and let me next think through what context I need to bring in, what kind of tools I need to add to this, to improve the experience.

— Why most AI products fail: Lessons from 50+ AI deployments at OpenAI, Google & Amazon, at 11:47

---

The biggest mistake companies make is believing they can jump straight to building fully autonomous agents on day one. Instead, successful AI products follow a deliberate progression that balances user control with AI capabilities.

Customer support example progression:

• V1 (High Control, Low Agency): AI suggests responses to human support agents who review and approve them, creating a feedback loop to understand where the AI works and where it has blind spots
• V2 (Medium Control, Medium Agency): AI responds directly to customers for certain categories of questions, while continuing to log human interventions
• V3 (Lower Control, Higher Agency): AI handles increasingly complex tasks like issuing refunds or raising feature requests with the engineering team

Coding assistant example:

• V1: Suggest inline completions and boilerplate snippets
• V2: Generate larger blocks like tests or refactors for humans to review
• V3: Apply changes and open pull requests autonomously

Marketing assistant example:

• V1: Draft emails or social copy
• V2: Build multi-step campaigns and run them
• V3: Launch, A/B test, and auto-optimize campaigns across channels

Kiriti Badam uses a helpful analogy: “Think about if your objective is to hike Half Dome, right? You don’t start hiking it every day, but you start training yourself in minor parts and then you slowly improve and then you go to the end goal.”