A Conversation with Tony Kipkemboi
Tony Kipkemboi grew up in Kaptagat, Kenya — a small town in the northwestern highlands that has produced an improbable concentration of the world’s greatest marathon runners. Eliud Kipchoge country.
A place where the morning runs start at 4am in the dark and the best runners in the world are your neighbors.
Tony wasn’t one of them. But he ran. And running was his door out.
The Non-Linear Path
A running scholarship brought him to Georgia, then to Grambling State for NCAA Division I cross-country. The Army’s language recruitment program brought him in during 2014. Seven years of service, most of it as a medical lab tech, ended at USAMRIID — the US Army’s infectious disease research lab in Maryland — in the middle of COVID.
“One of the first samples of the virus was sent to our lab,” Tony told me. “And our job was to aliquot each of those samples and distribute them to all the labs in the country so they could start working on vaccines. While everyone was quarantined, I was not.”
He discovered Python in that same lab, during a genomics rotation.
Spent months teaching himself to code on three hours of sleep a night. Earned a master’s from Penn. Went to Bloomberg, then Snowflake, then Crew AI at a thousand GitHub stars before anyone was talking about agents.
Now he’s at Guild.
Workflow First, Agent Second
Tony’s core philosophy for building agentic systems is simple:
“Workflow for me is the first citizen. And then the agent comes as the second layer. If you go even deeper, then there’s data that the agent is going to sit on top of to basically drive the logic. Most of the data was not built to be consumed by agents. They were built for humans.”
Most teams skip this.
They see an AI opportunity and jump straight to the agent layer, without mapping the underlying process, without cleaning the data, without understanding what they’re actually trying to solve. Tony calls this the magnifying glass problem: AI doesn’t fix broken data infrastructure, it exposes it.
The Real Signal
But what struck me most in our conversation wasn’t about agents at all.
I asked Tony how he knew he’d found the work that mattered to him.
“The way I figured out what I enjoy is the fact that even if I’m not working, I’m still living it. If you go to my phone, the last ten days of searches – it’s something involved in my work. I am obsessed with it. My social media feed, who I follow, is basically that. Every conversation I have with my friends – it’s basically that.”
“My off day is when I’m asleep.”
That’s alignment.
And it came not from a career plan but from a series of two-door decisions: if there are two options, just pick one and see. If it doesn’t work, go back and take the other door.
Watch the Full Conversation
We also talked about how to teach people to use AI as a thought partner instead of a crutch, why philosophy is becoming more important in the AI era, and why skills and community are the two anchors that work in any future – utopian or dystopian.
The full episode is embedded above (paid subscribers get early access to Practice Ground episodes).
Also on YouTube and Spotify/Apple Podcasts.
I hope you enjoy this conversation.
Nifemi
