Menu

I Sold My AI Company to Get My Real Job Back

How I pivoted from client-facing PM to deep-tech founder, and the strategic lessons learned on the path back to the customer.

I Sold My AI Company to Get My Real Job Back

There are moments that give you perfect, cinematic closure. Selling my first company was one. Selling my second was different. It wasn't closure. It was a ticket home.

The Discovery

Let’s rewind. The sale of Clear Review, our performance management platform, was a storybook ending. We’d carved out a space in the dense, brutal HR Tech landscape—a world dominated by giants like SAP and Workday—and built something valuable enough to get a life-changing offer. For me, the foundational PM, it was mission accomplished. A clean exit. Closure.

With the freedom that capital provides, I took on a new challenge: Knowledge Drive. The mission was born from over 200 Jobs-to-be-Done interviews. The insight was crystalline. The single biggest boulder blocking internal knowledge sharing wasn't a lack of platforms. It was the agonizing, manual, soul-crushing effort required to take unstructured data—the knowledge trapped in documents, presentations, and emails—and make it clean, accessible, and useful.

This was pre-ChatGPT. The peak of the art was libraries like spaCy and AllenNLP. Ripping intelligence from a PowerPoint slide was less a software problem and more a dark art. And in that moment, I discovered something fundamental: the product I needed to build wasn't for a user. Not yet. The first customer was the machine itself.

The Strategy

The strategy, then, had to be radical. It required me to reposition the most important asset in the company: the product manager. Me.

I had to stop being the customer-facing, workflow-obsessed PM I’d always been and become something else entirely. An R&D PM. A deep-tech PM.1 My focus shifted from user journey maps to dependency parsing trees. From stakeholder interviews to advanced Python. From designing a better dashboard to architecting a better brain.

It wasn't about finding product-market fit in the traditional sense. It was about achieving problem-technology fit. I had to go into the engine room, get my hands dirty, and build the engine myself.

This meant a full-immersion dive. I learned the intricacies of natural language processing, the linguistics of entity recognition, and the brutal elegance of building a backend architecture based on microservices.2 It forged an API-first mentality into my DNA. It was a detour, but a necessary one. You can’t build a house on sand, and you can’t build a knowledge engine on a weak technical foundation.

Execution in the Engine Room

For years, my product roadmap wasn't measured in features, but in technical milestones. We weren't iterating on a UI; we were iterating on an AI. Our beta customers weren't giving us feedback on button placement, but on the precision and recall of our extraction models built with tools like spaCy.3 It was a different language, a different cadence, a different world.

Then, the ground shifted. GPT-3, and later GPT-4, arrived. Suddenly, the world had access to the kind of power we’d been painstakingly building from scratch. A lesser product thinker might have seen it as an existential threat. I saw it as the ultimate validation—and a new opportunity.

This is where the old PM skills roared back to life. I knew from years of experience that for an enterprise, the shiny new tech is always secondary to the real-world anxieties. The number one question wasn't, “What can it do?” It was, “Is our proprietary data safe?” Our ability to pivot and build a solution using federated LLMs—keeping a client’s data secure within their own walls—was a direct result of product management discipline.4 We didn't just see the technology; we saw the trust gap. And we built the bridge.

The Payoff

That bridge became our most valuable asset. The deep, technical IP we’d developed was precisely what made Knowledge Drive an attractive acquisition target. The negotiation was keen, the offer was right, and it was time to sell.

But the real payoff wasn't just the exit. It was the transformation. I had gone into the wilderness a workflow specialist and emerged a full-stack product leader. I now understood the entire stack, from the customer’s unmet need all the way down to the microservice architecture that serves the API that powers the model that solves the problem.

And yet. I missed my old job. I missed the team. I missed the velocity of shipping user-facing improvements. I missed the direct, human-to-human collaboration of building something for people, not just for machines. The sale of Knowledge Drive wasn't just a financial success; it was a strategic decision to return to the arena I love most.

The Playbook: Lessons from the Detour

  1. Know Your PM Archetype, But Don't Be a Prisoner to It. Understand if your center of gravity is the user, the business, or the technology. But recognize that the problem itself may demand you step into a different role. True growth happens when you stretch beyond your archetype.
  2. Embrace the Necessary Detour. Sometimes the fastest path to the customer is through the machine. Don't be afraid to go deep, to learn the code, to understand the architecture. That technical fluency will become your strategic advantage later.
  3. Translate Technology into Trust. The most advanced AI in the world is useless if your customer doesn't trust it with their data. The product manager's job is to see past the technical capabilities and build for the human anxieties—security, privacy, and control.
  4. An Exit is a Strategy, Not Just a Goal. Selling a company can be more than a financial transaction. For me, it was a deliberate move to close a chapter of deep-tech invention and reopen a chapter of customer-centric creation, armed with a powerful new perspective.

Conclusion

They say you can’t go home again. I disagree. You can. But you shouldn't expect to arrive as the same person who left. I sold a company built on an AI engine to get back to the work that energizes me most: client-facing, workflow-based product management. I didn't just sell a company. I bought back my favorite job, with a hell of an upgrade.

  1. Reforge has a fantastic breakdown of the different PM archetypes. Knowing your 'type' isn't about boxing yourself in; it's about knowing your center of gravity so you can consciously and strategically move away from it when the problem demands. Required reading, in my opinion. Read it here. ↩
  2. If you're a product leader who wants to truly understand modern system design, you have to understand microservices. Martin Fowler's original piece is still the canonical text. It's dense, but it provides the 'why' behind the architecture that powers nearly every modern tech company. It's a must-read. ↩
  3. In the pre-LLM world, open-source libraries like spaCy were the bedrock of applied NLP. It's a masterclass in production-ready, high-performance software. Understanding tools like this gives you an appreciation for the layers of complexity that large language models now abstract away. Check out their philosophy. ↩
  4. Federated Learning was a concept pioneered by Google years ago, but its strategic importance has exploded in the generative AI era. It's the key to solving the enterprise paradox: how to leverage powerful models without sacrificing data privacy. This early Google AI blog post is a great primer on the core concept. See how it started. ↩