Nine Years of Arvension: What I Know Now

Nine years ago this month, I decided to leave a comfortable job and start a company. I was 28. I had watched enterprise software for years and I was certain that everyone was doing AI wrong. I was going to fix it. Looking back, I was both entirely right and completely naive.

I've never written something this personal in a company context. But 2026 feels like a moment to reflect honestly.

What I Got Right

I believed in timing before I understood timing.

In 2017, when I started, everyone else was building AI as a feature or a layer on top of existing software. I looked at enterprise operations and I saw something different: I saw systems that could be fundamentally rethought if you started with AI as a first principle instead of an afterthought.

That insight was correct. It took nine years to prove it, but it was correct. The companies winning today aren't the ones that bolted AI onto their legacy ERP systems. They're the ones that rethought operations from the ground up with AI in mind.

I also got the bar-raising part right. I decided early that I wasn't going to ship mediocre AI. I watched other founders get caught in the hype-then-hangover cycle. They'd ship an AI feature that was 80% good, users would be underwhelmed, and then they'd spend two years rebuilding the same feature. I decided to take longer and ship things that actually worked.

That cost me three years of momentum. We should have launched earlier. But when we did launch, people used our products. They weren't impressed by the AI—they were just satisfied that it worked reliably.

And I got the customer selection right. Early on, I made a decision to only work with customers who understood that operating with AI would require them to change how they operated. We turned away deals with companies that wanted AI to fix broken processes. We only worked with companies willing to rethink their operations. That felt like we were leaving money on the table. In retrospect, that decision saved us from years of failed implementations.

What I Got Wrong

I thought the bottleneck was technology. It wasn't.

I spent 2017-2020 building. I thought if I built the right architecture, the right models, the right systems, enterprise adoption would be obvious. Companies would see what was possible and move. That was naive.

The actual bottleneck was organizational. Companies didn't know how to operate with AI. Their processes were built around human decision-making. Their risk frameworks assumed human judgment. Their compliance was built on audit trails of human actions. Dropping an AI system into an organization built for humans wasn't an upgrade—it was chaos.

I had to learn that real enterprise adoption required helping customers rebuild their processes, not just giving them better technology.

I also got the hiring wrong for a long time. I hired engineers first because I am an engineer. We needed business people and operationalists way earlier than we brought them in. For years, our sales conversations were 90% technical and 10% business. We were explaining the capability instead of the value. We could have compressed the first five years into three if we had the right people explaining the problem to customers in the language they cared about.

I was too focused on general solutions and not focused enough on specific wins. I built infrastructure for any enterprise AI problem. What I should have done is crush one specific problem—order processing, or invoice handling, or supply chain planning—and then expand from there. Broad platforms are powerful but they're also slower to prove. I would do this differently now.

What Surprised Me

The longest sales cycles I've ever seen started when I thought we were done building. Customers would say "this is amazing" and then take nine months to actually deploy it. I learned the hard way that technical capability is maybe 30% of what determines adoption speed. The rest is change management, organizational alignment, and trust.

I was also surprised by which customers moved fast and which moved slow. Smaller companies moved faster. Large enterprises moved slower. That seems obvious in retrospect, but I had assumed larger customers would have more resources and discipline to move quickly. What they actually had was more process, more layers of approval, and more reasons to be cautious.

And I was surprised by how much of building a company is about communication. When we figured out how to explain what we were doing in clear, non-technical language, everything changed. We started attracting customers who were problem-focused instead of technology-focused. Our hiring improved. Our retention improved. Most of what I've learned in the last three years has been about language and clarity, not about technology.

The Hardest Calls

The decision to stay focused when we could have pivoted is the one I think about most.

Around 2021-2022, everyone was pivoting to generative AI. ChatGPT came out, VCs were throwing money at anyone who could use the word "GPT," and there were moments when I questioned whether we were too stubborn, too committed to the original vision.

I chose to stay focused on enterprise operations instead of chasing the generative AI wave. Some of my early advisors thought I was making a mistake. I was losing momentum in VC conversations. Our growth rate looked slow compared to companies riding the LLM hype.

But I was right. The generative AI wave produced a thousand companies that are now getting crushed by the reality that GUIs with chatbots don't solve enterprise problems. We were quietly solving enterprise problems. By 2024, when everyone realized that general AI needed to be operationalized before it mattered, we had nine years of operational knowledge that competitors couldn't replicate.

I also had to make the call to shut down product lines that weren't working. We built things that were technically impressive but weren't solving real customer problems. Shutting them down meant admitting that months of engineering work wasn't going to ship. That's hard for an engineer-founder. But it was the right call because it forced us to focus on what was actually moving the needle.

What I'd Tell My 2017 Self

I'd say: don't worry about the technology as much as you do. Build just enough technology to solve the problem. Spend way more time understanding the customer's actual operation and constraints.

I'd say: hire non-technical operators earlier. You think you're building a technical company. You're actually building a services and consulting company dressed up as a software company. That's not wrong—you just need to staff it accordingly.

I'd say: trust that if you solve the problem for one customer really well, other customers will find you. You don't need to sell every customer. You need to make one customer so successful that they tell everyone else what you did.

I'd say: the speed at which you build is not the speed at which the market moves. You will finish a feature in three months and it will take six months for customers to realize they need it. Plan for that.

And I'd say: you're going to have moments where you question whether any of this is real. You're going to worry that everyone else has figured something out that you haven't. They haven't. Most companies are confused about where AI fits into their operations. You're going to be one of the few that has a coherent answer. Trust that.

Looking at 2026

I don't know what the next nine years look like. But I know what the next 12-24 months look like: AI agents operating enterprise systems. Autonomously. At scale. Every company that has spent the last five years building AI as a layer will suddenly realize their architecture doesn't support agents. Every company that has spent five years understanding their actual operations will have a two-year head start.

I built Arvension for this moment. Not because I predicted it perfectly, but because I stayed focused on one question for nine years: what does enterprise operation look like when AI is not a feature but a participant?

That question has kept me sane and kept the company focused through hype cycles and skepticism and moments of doubt.

If you're starting something now, find your one question. The companies that will matter in 2035 are the ones asking the right question in 2026 and staying obsessed with it while everyone else is chasing the trend of the moment.

That's what I should have told myself nine years ago. And it's still true.