Sasha | Acepe.dev

- Agentic Development Environment with Orchestration

[Sasha | Acepe.dev @sasha_zelts]

Acepe's review system, takes the best ideas from Cursor's UX and making it agent agnostic, Claude Code, Codex, Opencode..

Always see the code changes your agent is making and understand them before calling it a day.

[Matt Shumer @mattshumer_]

I'm so angry... the OpenAI team is looking into it, but this feels like something that should happen with GPT-3.5.

Not a mid-2026 frontier model on the highest reasoning level.

[BridgeMind @bridgemindai]

GPT 5.6 Sol just hit CursorBench.

The economics are brutal for Anthropic.

Fable 5 Max: 70.5% at $17.32 per task. 103,525 tokens.
GPT 5.6 Sol Max: 67.2% at $5.22 per task. 28,320 tokens.

95% of the performance. A third of the cost. 73% fewer tokens.

Fable 5 kept the crown on raw score. It lost on everything that shows up on your invoice.

And remember: GPT 5.6 comes with limits you can actually live with.

The frontier is not an intelligence war anymore.

It is a value war.

[Sasha | Acepe.dev @sasha_zelts]

@Aquiverr @blknoiz06 amazing!

[Sasha | Acepe.dev @sasha_zelts]

Mobile is buggy, it messes with the permissions mode (meaning if you control from mobile, it will sometimes start asking for permissions.
Sometimes you try to stop a chat , but it doesn’t work. And I will be spam clicking it and just nothing responds.
I would say it is inferior to Claude Remote Control which feels smoother

[Sasha | Acepe.dev @sasha_zelts]

@wongmjane @Engineering @X That was long due man

[Sasha | Acepe.dev @sasha_zelts]

@robinebers Ain’t hodling CBRS for nothing.. really believing in those lads.

[Sasha | Acepe.dev @sasha_zelts]

@tannerlinsley @shadcn love to see this kind collaboration between top of the notch libraries.

[Sasha | Acepe.dev @sasha_zelts]

Turns out the safest spot on the frontier is one step behind it.

Smart move @sama

[OpenAI @OpenAI]

We believe in broad access and plan to make GPT-5.6 Sol, Terra, and Luna generally available in the coming weeks.

For now, at the request of the U.S. government, we’re starting with a limited preview among a small group of trusted partners in Codex and the API.

[Sasha | Acepe.dev @sasha_zelts]

Great to see Japan in the race as I have huge respect for their products in general

[Sakana AI @SakanaAILabs]

Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API.

Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls.

Try it: https://t.co/hhO6qTawgb 🐡

[Sasha | Acepe.dev @sasha_zelts]

@mickces what if you have multiple sessions?

[Sasha | Acepe.dev @sasha_zelts]

@lexrus Nice, whats the app btw ?

[Sasha | Acepe.dev @sasha_zelts]

@afshineemrani I’m curious, as someone not in the medical world , and sure you can say that with most innovation, but why has the medical world not thought of this earlier ? Or was there a technological constraint? Something got invented recently? Besides LLMs?

[Sasha | Acepe.dev @sasha_zelts]

@TheWorstGameDev @MakerMatters @gdb No you just have a hook that reminds the agent then he queries the memory on relevant keywords and gets insights about what to do what went wrong before etc

[Sasha | Acepe.dev @sasha_zelts]

@MakerMatters @gdb Same man, I noticed and it’s crazy slow. Adding a memory layer helps though.

[Sasha | Acepe.dev @sasha_zelts]

@impeccable_ai really interesting approach, gonna test it out.

[Sasha | Acepe.dev @sasha_zelts]

@mattpocockuk So you implement a plan that wasn’t reviewed ?

[Sasha | Acepe.dev @sasha_zelts]

I agree, until the point were we are building autonomous system which re-architecture their own code themselves. Oh wait, we already are.

And that's when you need quantification. There won't be pilots anymore but the agents/llms themselves. We will be writing the skills (or not?) and that's were I feel evals will be the determining factors.

An agent orchestration system needs to create a landing page, will he use 5.5 ? No because 4.8/fable 5 is the designer that the consensus said was better. But that one is easy to tell apart.

Code Architecture is a more complicated subject, but that doesn't mean we shouldn't think about it, and think we will always be the ones thinking of it. You know what I mean?

[Sasha | Acepe.dev @sasha_zelts]

I miss @cursor_ai tabs days. It just felt different. You felt in control. But hard truth is that it is so suboptimal right now.

But, I would love to see a "Cursor Tab Hackaton" event, where people still need to think through ideas entirely by themselves like the good old days, and get helped by tabs for the implementation. No agentic help whatsoever.

[Sasha | Acepe.dev @sasha_zelts]

I meant how do you quantify if one architecture improvement made is "better" than the other?

Performance improvements, scalability, extendability, use of programming patterns, etc

I meant more in that sense.

Software is about providing something right, for instance a software as a service provides a service to customers. We want that service to be a seamless experience, and that seamless experience depends on hundreds of factors.

That's the beauty of software, there is no single solution, but some softwares are way way seamless than the others, and that's thanks to their code (meaning architecture).

So my reflection was more about how do you quantify that. But I guess its kind of empirical at this point, which is nice, cuz our expertise (us as in AI engineers...?) cannot be reproduced so easily.