AI & Implementation
-
I Deleted 70,000 Comments
On 18 August I deleted every comment in a 29-day-old codebase: 71,660 lines across 742 files. Agents wrote most of them, and agents believed all of them. Here is where the reasons live now, what broke, and how the comments came back.
-
The Atoms Are Already Deployed
Cloud AI sells itself as a software business. Underneath the API sits the largest industrial buildout in the history of computing, and you are paying the depreciation by the token. The hardware that matters is already on your desk.
-
One server, every device, yours
Whelk is out. One server on hardware you own, a native client on every device you carry, and a room full of people running AI without a meter.
-
The future of AI is private
Two curves are crossing. Model quality is a wasting edge and hardware is sprinting the other way. When they cross for good, intelligence stops being a subscription and becomes a utility you own. Here is the bet behind Whelk.
AI & Implementation · 4 min -
The Depreciating Asset
Most AI products invest in the part of the stack that loses value fastest. The model is the wasting asset. The interface, the coordination layer, and the trust properties of where inference runs are what compound.
-
I Gave My Agents a Memory That Can’t Leak
My agents forgot everything when a session ended. So I built commonplace, an open source memory with two tiers: a hosted model reads my own notes, and client work under NDA never leaves the box in my office.
AI & Implementation · 4 min -
The Org Chart Is the Product
For most of my career I thought I was building products. Looking back, I was mostly building org charts. The customer gets the coordinated output of the whole company, and for the first time that coordination is something you can build instead of staff.
AI & Implementation · 3 min -
Polymorphic Cultures: One Year In
I have said polymorphic cultures for a year without writing down what I mean. Here is the definition: a team where a human, an agent, or both can do a unit of work, and the culture holds whichever one shows up.
AI & Implementation · 7 min -
The Marginal Return of Intelligence (And the Marginal Return to It)
People ask how I plan to build more than one company at once. The question assumes intelligence is still scarce. It is cheap now, and judgment is not. Two curves explain the gap, and most people fold them into one.
-
One-Person Companies Are a Forcing Function, Not a Fad
A one-person company at a million dollars a year is not the interesting part. The interesting part is what had to change for it to exist. Time, capital, and headcount set the floor for twenty years. The floor moved.
AI & Implementation · 4 min