Calling per token usage of the US closed source labs has always been funny to me, we have a clear model of what it actually costs to host these models from open models.
Your subscription is not subsidised, it is just closer to the actual cost of the model…
The tipping point from steady improvement to steady enshittification is just post the 2008 recession. As soon as the financial incentives were put in for growth (2009+) the trend shifted hard to value extraction from a captive audience.
In my opinion Wikipedia is worse than no Wikipedia. Wikipedia played a big part in suppressing what the Web might have been and started out becoming before it changed direction.
I was watching in real time as Gopher, Wais, and the Web were all equally in play, and as the Web won. I had a personal website at the beginning, and I still have it. It seemed to be the next revolution in the dissemination of ideas after Gutenberg. Individuals could make their creations available to the entire world at almost no cost. Individual sites, individual voices, in all their eccentric glory, arose in profusion. Then it all became swamped by anonymous, unaccountable corporate voices. Wikipedia is a salient example of a collective, bland, plagiarising substitute for individual expertise that can not be held responsible for its words and that is accorded a false authority. it is part of the tragedy of the Web that could have been.
I remember the time before smartphones. They make many tasks faster, easier, many tasks possible at all and replace specialised devices. You are wrong.
Markdown instructions are useful as durable context, but they do not replace programming. Their job is to keep intent out of a transient chat so both the developer and agent can revisit it. The actual constraints still belong in code: types, tests, linters, and dependency rules. Natural language can explain why a boundary matters; executable checks determine whether a change respects it. That feels more like a persistent layer around the programming workflow than a new programming language.
I agree. I keep agent memory off. In my experience, it remembers the wrong details and applies them in the wrong places. Project context should live in the repo, where the whole team can find it, review it, and correct it.
It should be treated as an explicit artefact of the codebase for Humans and Agents to work with.
I’ve never felt comfortable with how people use Claude Code or Codex as synonyms for Harness. It’s not wrong, it just reduces the harness to the least interesting part: the runtime. It’s the part you have the least control over. Most of us use what ever subscription makes most sense or whatever our company has settled on.
To me the interesting part of the harness is what you add on top, Brigitte Böckeler calls it the outer harness. I prefer the term Your Harness, your harness breaks down into two things: Skills and Loops.
Skills change what the agent knows and the loop changes how it works and checks itself. Both can mostly be encoded in Agent Skills
Your Harness should mostly agnostic to the agent you use.
