manmal

Seems to me a bit insensitive or logarithmic. Fable is way worse than some of the others in this list, but only 10-20% higher score.

There‘s a lot of ancient biology at play in raising kids. You won’t get exactly this kind of feeling elsewhere. Doesn’t mean it has to be your kids though. You just need to be an important part of their life.

Why wouldn’t you use Luna for that? It’s super cheap.

Yeah. Local agent sessions are not backed up in the cloud. And uptime is better with local models.

Isn’t the fact Fable is more expensive than Sol-Max by multiples already an indication that Sol is way smaller?

Are you using the SOTA models at very high reasoning during planning? IME that makes a LOT of a difference. I‘d also never let them just rip into the architecture, but always push back and ask for alternatives first. Once the overall plan is nailed, not that much can go wrong. Provided it’s a reasonable change set and not a 20k LOC PR.

I do spend 5x more tokens on planning and reviewing, than for implementation. But architecture is still nothing I can delegate.

> At a certain speed point, you're able to move from request -> response to a cascade of tool calling and "subagents"

That also needs server class hardware though. A phone won’t happily service the insane amount of IO, compute, and network that this cascade would require.

I think that only works if you have checkpoints where all that reasoning can be checked against reality. Otherwise you get an army of armchair experts. LLMs are hilariously bad at home improvement advice btw, where reasoning alone won’t get you far.

About your first example, isn’t the butterfly effect preventing this from being useful? One agent in your simulation decides to sell, and starts an avalanche, that won’t happen in reality?

Problem is, there exists no judge model that will really pick the same winner that you would.

Errors compound, and making 1000 wrong decisions per hour, will not result in something useful. Maybe you‘ve tried setting up guardrails for good design or architecture at some point? I think it’s simply not possible to do that.

It would certainly be an accelerator for people who know exactly what they want. And it would remove multi tasking, which I‘d appreciate.

That’s their API with extra steps, or am I missing something? That was always faster.

At this point I'm happy when things don't break randomly. However, it seems I lost an ACP session today and I'm not sure it's my own fault. Could be Zed playing tricks.

It’s not out of their domain at all IMO. I‘m only using Zed now (despite its non existent GH integration) because it works with Codex.

What I really like is that all code links it outputs, open a pane at that line.

I‘m using codex no issues, over ACP.

I use it constantly but haven’t felt the need to paste an image yet. I guess I‘m using pi and the Codex app more for that.

Their ACP integration is good (finally), at least with Codex. For when I want to be near the code, I prefer it now over TUIs or the Codex app. It feels much more immediate that way. Sticking to the basics will kill them IMO. They have been pitching themselves as agent-native for a while btw.

And the workflow they are describing in the post sounds amazing, I‘ve been wishing for that. It‘s orthogonal to CVS, no? You can still make PRs. But, crucially, you can loop people in early on, when it matters the most.

As a heavy pi user - this doesn’t help, for codex. The thinking traces are encrypted and you cannot move that session over to eg your local ds4. It will lack those precious thinking traces that you absolutely need to continue a session productively.

Yes there was a generation with plastic back which fell apart. Everybody uses double sided glass now. Should be good.

Sounds like the kind of project that would have afforded a Vision Pro, right? I bet you can spot mistakes before building it, that way. Things like, oh, this room will be very dark.

Or, good that the craze for ever thinner laptops has been slowed?

Panels should do 30 years, easy. Inverters will usually fail after 15y, and with batteries it’s unknown what the current gen will do - 15y, maybe 20? Definitely get LFP.

Have you tried a virtual board, like Miro?

It would take longer to train on intercepted Fable data, no?

Read the session file of the reviewer tomorrow. Sonnet should suffice.

Fighting against the weights is often a lost cause. For any model.

> The best agentic developers are now probably exceeding 100x, doing massive rewrites of codebases that would have taken years of engineering in days

I‘m not denying that you can get far with a port of a well tested codebase. But it’s a bit of a selective example, no? Porting a large framework to Rust is not something that’s making up a meaningful fraction of a developer’s time usually. It’s also a bit of a luxury IMO, and something you could have skipped entirely.

The trend to put acoustic panels or an imitation of their pattern on _everything_ is one of my pet peeves. They create some stroboscobe effect with my (not normal) vision. The result is that they mess with my 3D perception and break my brain a bit for that reason. I’d rather not have them put anywhere.