Observations on Technology, Business and Innovation
Two point one trillion dollars. That is the “backlog” the hyperscalers book as future revenue, and The Second Derivative: Why No One Understands the AI Boom shows what it really is: a concentrated, unsecured loan to two labs that live on their next equity check. The essay watches not the level of AI spending but … Continue reading Tokens Are a Substrate, Not a Market
So you vibed up a mess. Weeks of conversation, something real at the end of it, a browser tab you will not close, and a laptop you have quietly stopped rebooting. I know the shape of it. I once looked up from my phone to find a whole PID fan controller sitting in a chat … Continue reading Turning Pro in an Afternoon
RFPs, product specs, discovery, inheriting a codebase — formalizing what good PMs, founders, and engineers already do, plus a notebook that keeps the reasoning alive. In all my years, the phrase I feared most (other than “I’m from the government, and I’m here to help you”) was how hard would it be to… It never came … Continue reading How Hard Would It Be: A Working Pattern for the Start of Any New…
Andrej Karpathy’s LLM Wiki pattern — call it a KWiki — is one of the better ideas to come out of the agentic-coding moment. The premise is that an LLM should not have to rediscover your knowledge from scratch on every question. Instead, the agent compiles raw inputs (transcripts, papers, decisions, meeting notes) into a … Continue reading A Karpathy Wiki for Projects Already in Flight
An OpenClaw failure mode, and the CQRS-shaped pattern that replaces it If your agent reads a structured file, modifies it inside the language model, and writes it back, you are running a latent data-loss bug. This pattern covers any file that accumulates state across sessions: task queues, contact lists, memory stores, ticket backlogs, time logs, … Continue reading How the LLM Ate Its Own Homework
I was finishing the retrospective when the sponsor asked for the headline number. I told her: 11x. Lines of code per person-hour, against the traditional construction estimate. Measured clean, against a real baseline, on comparable scope. Her eyes lit up. She did division in her head. A seven-figure engagement, compressed by 11x. Somewhere a sales … Continue reading The 11x Number and the Software…
In 1999, a race condition surfaced aboard NASA’s Deep Space 1 spacecraft, 100 million miles from Earth. No ground testing had revealed the bug. Engineers opened a Read-Eval-Print Loop on the spacecraft itself, a live interactive session running Common Lisp relayed through the Deep Space Network, each command-and-response round trip taking over an hour at the speed … Continue reading Are We Not…
This is a work of fiction built to illustrate four points about how organizations adopt AI. The first is that the human-AI team is the productive unit, not the human or the AI alone. A machinist and a CNC machine transform aluminum billets into aerospace parts. A developer and an AI tool transform requirements into … Continue reading Fear and Loathing of Token Limits: Introduction
The best hours happen before the building wakes up. 6:30 at an enterprise software company is the temporal equivalent of owning the place. Security guard nods. Coffee machine is mine. HVAC hums its white-noise lullaby and every footstep on the tile sounds like a declaration of intent, and the coffee is the exact temperature where … Continue reading They Never Ration the Aluminum
The error handling work is going better than the morning session. Shouldn’t be possible, but is. Deep in the retry logic for the message transformation layer. The morning prototype handles the happy path: purchase orders conforming to the twelve most common patterns, transforming cleanly from legacy XML into the AI layer’s JSON and back. But … Continue reading The Wall