"Running ultras. Running companies. POV from races most won't run."
The southernmost ultra in the world. Tierra del Fuego. Cold, technical, unforgiving terrain at the edge of the continent.
At km 65, my knee collapsed. Not a dull ache. A structural failure. The race should have ended there.
I kept running. On painkillers. For 25 more kilometers. I crossed the finish at km 90 — not because I'm tougher than the mountain, but because stopping didn't feel like an option I'd considered.
That's not a heroic story. That's just an honest one. I didn't win. I didn't even finish the full distance. But I moved forward when moving forward was the only thing I could control.
I think about km 65 constantly. In boardrooms, in agent design sessions, in every moment a company problem looks like it should end the project. It usually doesn't.
The Pyrenees. July 2026. 110 kilometers through high-altitude terrain across the border of Spain and France.
I'm training for it now. My knee is better — mostly. I don't know if I'll finish. That's not false modesty. That's the honest condition of anyone entering a race this length.
What I know: I'll start. I'll move. And whatever happens at km 65 this time, I'll know what I did the last time someone asked me to quit.
Ultra running and building companies with AI agents are the same problem in different terrain.
In a 130K race, you don't have perfect information. You don't know how your body will feel at km 80. You don't know if the weather holds. You train, you show up, and you adapt. The plan is a scaffold — the race is the reality.
Building with AI agents is identical. You design systems, you specify outputs, you deploy. Then reality diverges from the spec. The agent does something unexpected. The workflow breaks at 11pm. A donor lead data source changes format. You adapt.
Most people treat discomfort as a signal to stop. In ultras, discomfort is just the baseline condition. It doesn't mean danger. It means you're running. The ability to distinguish between "this is hard" and "this is actually over" — that's the skill. In races. In companies. In AI deployments.
The honest number: $120/month of infrastructure to run three companies. Not because it's cheap — because agents handle what would otherwise require headcount. But getting there took suffering through broken workflows, bad specifications, and a lot of km 65 moments where it would have been easier to conclude "AI isn't ready yet."
It was ready. My specs weren't. That's the lesson. That's both lessons, actually.
The philosophical spine that connects both worlds.
At km 65, my knee wasn't controllable. My pace, my direction, my decision to keep moving — those were. In companies: the market, competitors, and AI capability curves aren't controllable. Your specifications, your systems, your response to failure — those are.
Pain tells you something is happening. It doesn't tell you to stop. The Stoics called this the discipline of desire — learning to strip sensation of its narrative. "This hurts" is data. "I should quit" is a story you added. Know the difference.
In the last 25 kilometers at Ushuaia, I didn't feel like running. I chose to run, and then eventually it felt like the right thing again. Same with every broken AI deployment at 2am. You choose to fix it before you feel like fixing it.
I didn't finish Ushuaia 130K. I finished km 90. The honest version of that story is more useful than a cleaner one. Same principle applies to agent deployments: "We're using AI" and "Our agents automate 40% of this workflow reliably and break on edge cases we're still mapping" are different statements. Say the second one.
Ed Endures. · @ed.endures · edendures.com · Buenos Aires