Bob McWhirter: Don’t Ship Infinite Monkeys
Bob McWhirter argues that open-ended agent exploration belongs in development, while shipped AI features must be bounded and proven on utility, reliability, budget and failure behavior.
Bob McWhirter argues that open-ended agent exploration belongs in development, while shipped AI features must be bounded and proven on utility, reliability, budget and failure behavior.
Compute power can reproduce almost anything; it can't imagine anything. "Learn to code" pointed people at what AI is able to automate, and we should have been asking for imagination, not lines of code.
Information is only scarce when somebody makes it so, and the people who refuse to make it scarce are the ones rewarded least. They are carrying a load they can’t bear forever.
In "The Geometric Model of Maintenance," someone finally checked the “one month of code, ten days of maintenance” rule against real git history. The number holds up. What it’s made of is worth a look, and so is what you’d need to measure before it tells you anything about your own code.
There's no standard for marking AI contribution, and there can't be, because contribution isn't a scalar. What readers actually want is attestation, and that's what a byline always was. The question was never whether a machine touched the text. It's whether anyone was behind the controls.
Baldur Bjarnason argues that development tools win by making developers replaceable. The JVM ran that experiment three times, against Visual Basic's precedent, through EJB's committee, and into Spring's reversal, and the results say his theory is exactly half of a whole.
We've decided, collectively, that the things we love should be free and paid for by someone else. Thunderbird just put a number on how that's working out: 3% of their users actually invest in the project. We want good software, honest journalism, and open source tools that last, but we just don't want to be the ones paying for them. That's a problem.