My curiosity for AI started in 2015 when I was part of the Fast Ventures team at Deloitte. AI was one of the frontier technologies slowly making its way into the enterprise and we were advising clients on which ventures to acquire, invest in, or partner with. I spent hundreds of hours in Crunchbase and Pitchbook researching AI scale-ups. I got hooked and decided I wanted to work at one.

In 2019 I joined Tamr in Cambridge, Massachusetts, just off Harvard Square. Michael Stonebraker and Andy Palmer started it out of MIT CSAIL to solve a problem most enterprises were stuck on: making sense of messy, duplicated data across systems. Tamr pioneered the use of ML-based entity resolution, and companies like Bank of America, Toyota, and Thermo Fisher were using it to unify their most critical data. This was the first big AI wave for the enterprise: prediction models.

In November 2022 OpenAI released ChatGPT to the public and we entered the second wave: generative models. LLMs keep getting better at creating text, images, and video. If you're using Claude or ChatGPT daily, you feel the difference. Hard to go back.

Fast forward to today, and we're at the start of what I think will be the biggest wave yet. Agentic AI. Systems that don't just generate outputs but actually do things across multiple tools, chaining capabilities together, and handling exceptions. At 5CA, which runs outsourced customer support for some of the biggest gaming and consumer brands, a big part of my job was helping build the foundation for exactly this: AI that doesn't just assist agents but runs workflows end to end.

These waves don't replace each other, they stack. Most organizations haven't finished wave one. They're still wrestling with data quality while trying to bolt on generative and agentic capabilities at the same time. That's the reality we work with.

I've decided to go all in on this, launching Third Vector with two focus areas:

1. Going AI-native: helping businesses redesign operations across the full AI stack, from data foundations to agentic workflows

2. Building AI-native: partnering with companies creating the capabilities, from domain agents to orchestration layers

Our first partnership is with 5CA, built on 27 years of running customer support for the biggest names in gaming in a tech-driven way. 5CA developed AICX: the orchestration layer that sits between your support team and your tools, and that's exactly where agentic needs to work.

There's a lot that still needs figuring out. How do you actually rewire operations instead of just adding tools? How do you deal with bad data when the system is now acting on it, not just displaying it? Those are the questions I want to spend my time on. If you're working through the same challenges, let's talk.

We'll be writing about what we're learning on Substack, together with co-authors who are building right now.