"indifferent" was the word they used. The sensible way to use Fable and Opus to do work at this point is claude code, but that's fine. I always have several instances running, along with codexes and pis. The question to ask is: can you stop a claude code and start a codex in its place without disrupting the team? Does my infrastructure support learning in a way that is stored by me, and not tied to a provider?
I built a harness that externalizes state into files so you can swap providers. I regularly move between claude and gpt, but it works with all providers including pi and local models.
I log everything - user messages, tasks, project memory, even bash commands for forensics. As a consequence you can do reflection where you analyze past work and extract refinements for the harness and realign the project when it diverged from user intentions. I don't have to do this manually, it reduces steering work.
It's cheap, just add a line to .bashrc, every time bash is invoked this script is executed and appends a line to a file. I rarely need to see bash_history except when I suspect agents did a boo boo.
It might be against ToS, but as a pragmatic matter, Anthropic doesn't seem to care if you use a custom harness with a subscription atm. I know a lot of people doing that now. Using Opus 5.5 as the orchestrator, and Luna max as the workhorse.
I’ve rolled my own harness in Go. I have a rule to not let the production LoC exceed 50k lines. It is _very_ nice for my use cases. DeepSeek Flash V4.1 often performs at the level of GPT 5.6 Sol for non-orchestration tasks (programming and maths).
I add features and tweak it all the time: You just need to get comfortable with spending 3 hours in a chat session every 3 weeks to think hard about how to simplify whatever slop you didn’t simplify the last time you did this :)
As a programmer, maths researcher, and painter, there’s certainly something satisfying (to me) when an artistically symmetrical piece observes no mathematical symmetry: The very asymmetry in the language used to describe it is, in and of itself, beautiful.
Not sure what you mean by "an artistically symmetrical piece observes no mathematical symmetry" - as in, the underlying structure has a (set of) symmetries but the 'decoration' on top is asymmetric?
Only under gross negligence would it be unprovable: Did you use a model whose training set included user data? Did the transcripts of any of the agents include a tool call whose result including user data?
>Did you use a model whose training set included user data?
OpenAI, as with all AI companies, openly admits that it trains on user data unless the user opts out. But the mathematicians have not said whether or not they opted out.
>Did the transcripts of any of the agents include a tool call whose result including user data?
>>Did the transcripts of any of the agents include a tool call whose result including user data?
> They have already explicitly denied this.
AFAICT they only denied accessing data through a request targeting a user, not that they accessed (users) data through targeting (an extremely niche) topic, which is the relevant part here.
So, unless I'm mistaken, a "result including user data" is very much still in the air.
I only made it through the first round of prompt selection; both options for the second step were equally pointless and not at all prompts I would ever expect to result in a constructive outcome. In my experience, telling the model it screwed up without specifically addressing, unambiguously, how to fix it, only leads to more suffering. If this page illustrates nothing else, I think it shows the immense downside of trying to use simple one or two sentence prompts.
EDIT: Actually, I used to use Google's AI Studio a lot and fork it after every successful prompt interaction. When I'd encounter a problematic issue like this, I'd revert to the previous fork and try a different prompt until I could get the desired outcome, thus mitigating the need to "argue" with the LLM. Unfortunately the ability to cleanly fork and revert everything including the LLM context was removed some months ago, and I've yet to discover a workflow with any tool that works as well for me.
> In my experience, telling the model it screwed up without specifically addressing, unambiguously, how to fix it, only leads to more suffering.
I wonder if this is just a reflection of some senior folks being arrogant towards junior folks. When the latter finished a task but not to the liking of the senior person they might just get a "that's wrong, try again". Just to have sth similar repeat the second time around. But the arrogant guy got to boss around the junior one, and some junior folks grow up learning that's how you should behave so they also do it later.
Now it's not a person but a machine. And people just make fun of the dumb machine. Well, garbage in, garbage out If you are not specific in what you want, you might get crap back. Or at least sth you didn't envision.
How is "Make the shopping button blue" not specific? Any human with a lick of digital common sense would be able to understand what I obviously mean. How am I supposed to anticipate every single way the AI could possibly misinterpret a simple fucking instruction like that?
Absolutely, the initial prompt appears sufficient. But this is about the thing screwing up and the user giving corrective instructions. Saying essentially "are you dumb" is not helpful for neither a human nor a machine if you think the original instructions should have been sufficient. Clearly they were not in that case, if you like it or not. Just repeating or being condescending is not going to help.
Seriously: I roll my own harness over OpenRouter and GPT 5.6, and it is genuinely better than CC+Fable for my maths and programming use cases.
The one thing Fable 5.1 still tops out on is the ability to take a simple "Let's engage in an interactive complexity-removing session" and materialize abstractions and semantics quickly.
And, no, wrapping claude -p or any other “allowed” use wherein you don’t control the agent loop isn’t the same.
reply