parents: "move out"
girlfriend: “quit being such a loser”
boss: "work harder"
claude: "uber for dogs (the dogs are the drivers) is a great idea, you should absolutely pursue it"
You are employable to the extent that you can catch an agent's mistake...
Employees train the agent.
Owners own the configuration.
Every human's job is to create a system that teaches an agent.
Every agent config change will be a guaranteed improvement with back testing and evals.
The loop:
observe the output -> catch the agent's mistake -> add a regression test -> re-run the task.
Using Claude Code has a weird side effect: You don't just get more productive, you actually want to work more.
There's something addictive about watching a product being born in real time in front of your eyes.
"One last feature" after "one last feature" and it's already past 3am.
The choke point for iteration speed switched from how fast can we build it to how fast can we make decisions.
"Qualified" now means "I trust your judgement".
In the AI age skills are apps. Yet everyone gives their skills away for free.
Skills need to be executable in a sandboxed environment where the author protects their IP and the customer (user of the skill) has assurances their data used in context doesn’t leak.
The only moats left in software are network effects, distribution and brand.
Brand trust matters most when apps hold sensitive data (thus high switching costs in fintech).
First they came for the programmers and I said nothing because I was not a programmer.
Then they came for the Excel monkeys and I said nothing because I was not an Excel monkeys.
Then they came for the mathematicians and I said nothing because I was not a mathematician.
Then they came for the mechanism designers and there was nobody left to stand up for us.