A rule with no enforcement is a wish.
[Read More]
[Read More]
Essays by Frank Chen, co-founder and CTO of Balto Energy, on AI agents, knowledge systems, developer infrastructure, and behavior design — what actually ships vs. what demos well.
Blue-Collar AI Is an Observability Problem
The hard part is ground truth, and ground truth clocks in with the crew.
One year ago, Andrew didn’t have the words. Not in the fake-humble way. He’s the CEO of the operations company we’re forward-deployed with, and back then, literally: ontology was jargon, observability was a software team’s problem, and AI inside a residential energy contractor sounded like a demo from another universe....
[Read More]
[Read More]
FD(x): Forward Deploy to Solve the Untrainable
The cleanest number in our dataset was a lie, and only the operator who owns the loop knew why.
There’s a field in a public solar dataset that flags whether a system is a lease instead of a purchase. For our operations company it reads a clean, machine-precise zero. The data says, flatly, that the company has never once done a lease.
[Read More]
[Read More]
What It Feels Like to Rebuild a Company Around a Graph
Our CEO went forward-deployed into the ontology. The wonder and the friction, honestly.
“Why are French restaurants so good? Because the waiter is part of the kitchen: he knows the food, the technique, the method, not just the path from the pass to the table.” — the question Alex Karp asked Shyam Sankar in 2006, which named the forward-deployed engineer a year later...
[Read More]
[Read More]
The Handoff Was the Bug. Vertical AI Is the Rewrite.
Quality used to come from approvals. Now it comes from correction.
The Operations CEO Doesn't Wait for the Dashboard
What happens when the forward-deployed engineer is the CEO himself.
Seven a.m. Tuesday. Andrew, the CEO of the operations company we’re forward-deployed with, is at his desk, running a forecast iteration on a cost model he wrote himself. By the time most of his team has logged on, he’s already cycled through three or four iterations and pinged his sales...
[Read More]
[Read More]
The Moat Is What You Capture
Reasoning gets cheaper. Capture compounds.
By the end of the Palantir kickoff build week I’d figured out where the leverage actually is in a vertical AI build. Not the model. Not the agents on top. The graph underneath — and the loop that thickens it.
[Read More]
[Read More]
We Built the Enterprise Brain. It Was the Easy Part.
Fat skills, one per work function.
Alex Lieberman thinks someone is about to build a world-class
brain for enterprises and make a stupid amount of money.
He’s right about the money. I think he’s wrong about what
you’re buying.
[Read More]
[Read More]
Our AI Onboards New Hires Better Than We Do
How we use Claude Code to personalize onboarding from a living knowledge base
Last week I wrote about keeping our company’s brain in a
git repo. A couple founder friends DM’d me after: “this
sounds great, how do I get this started for my team?”
[Read More]
[Read More]
We Put Our Entire Company's Brain in a Git Repo
ARKS, AI agents, and what happens when your company's knowledge lives in plain markdown
+----------------------------------------------------------+
| We put our entire company's brain in a git repo |
| |
| I'm not sure it's the right move but I think |
| something interesting is happening |
+----------------------------------------------------------+
[Read More]
[Read More]