Before: "Don't spend too long documenting things so people can know what's going on, write code."
After: "Don't spend too long writing code, document things so the agents know what's going on."

Before: "Don't spend too long trying to set up the culture and processes of the team so the develop/test/ship loop is smooth, write code."
After: "Don't spend too long writing code, make sure the agents are in a smooth develop/test/ship loop."

We weren't allowed to do it for humans, we _must_ do it for the robots.

@benno GPUs have no history of unionising.
@benno turns out I have been thinking about this a lot https://social.coop/@chrisjrn/116121133780551153
@chrisjrn I think we've been thinking several of the same thinks for a while now.
@benno @chrisjrn hmm yea hey what's up with that anyway, is there some kind of seismic shift in the material conditions of the world or something going on 🤔
@SnoopJ @chrisjrn I dunno, I'm sure the economically and socially sustainable cottage industry of fabric weaving will be fine.
@benno @SnoopJ @chrisjrn I stand by my statements at PyConAU 2025.
@benno @rjbs Last May I speculated: “If the criticism is ‘that LLM stuff doesn't work unless you keep the code extremely modular’ that's not much of a criticism. We all need more encouragement to keep the code modular.… What if writing closely-coupled modules had an immediate cost today, the cost being that the LLM would be less helpful and more likely to mess up today's code? Maybe programmers would be more careful about letting that happen!”
@mjd @benno I've said many times in the last 1-2wk: "If a practice would help a junior dev, it would probably help an agent." And it's nice that this motivation exists. But it gives me feelings about how it's higher priority now for automata than it was for human apprentices. 🤷
@rjbs @mjd @benno because if the LLM can do the work, they can get rid of the human, which bosses have been excited for since forever.

@ainmosni @rjbs @benno Yes, and there used to be an army of guys (“longshoremen”) who made a living carrying cargo onto and off of ships down at the port. When cargo shipping was containerized starting in the 1960s, those guys were put out of work and now the profession barely exists. Bad for them in the short run, but in the longer run the whole world reaped the huge benefits of containerized shipping, even the families of the former longshoremen.

The problem that should be solved here is not container shipping, it's our social and economic system that allows managers to discard workers as trash if their skills become obsolete, and which don't provide any support to discarded workers.

The longshoremen got the shaft, and this is a real problem that caused real suffering. The right solution to this real problem would not have been to forego container shipping, which has brought wealth to literally billions of people. A good solution would have been to have better protections for displaced longshoremen.

Longshoremen were able to avoid some of the worst of the effects because they had a powerful union that had negotiated good contracts for them. Programmers haven't have so much foresight. But *if* programmers can be fired because the LLM can do their work, then everyone will reap the benefits, even programmers and their families. Again the proper solution isn't to forego the benefits of the new technology, it's to provide better support for the people with newly-obsolete skills.

@mjd @ainmosni @rjbs @benno that all might be true, but it seems most likely that we’ll keep the new technology without the new protections. (not to mention the other issues with LLMs besides worker displacement.)

@benno I've had the same sort of weird feelings about all this MCP stuff, which is basically just AppleEvents, Amiga ARexx message ports, CORBA, DCOM, and every other "we will define a protocol more structured than shell pipelines so you can link your apps together and Do Things" system we've ever had, only sloppier.

On the one hand, we've tried this before and gave up because it was too much effort. On the other if we end up finally doing it right because AI... dunno what that says about us.

@benno
Indeed! It's been rather disturbing seeing how some of these agents.md / context prompt things are put together, as a lot of them end up just being really good documentation to people on how they should be doing their jobs...

It's almost as if we didnt do it before to make it just slightly harder for management to roll in new warm bodies, whereas now, folks are doing it as they can't wait to obsolete themselves!

Reminds me how certain enterprise patterns + design decisions suddenly start looking very appealing for paid-by-the-hour contractors, but will be the first thing to go if you've got a small team under the pump to get things ddelivered within tight timeframes!