Beshr

System Engineer / Architect.

[Beshr @beshr]

@yuvadm Yeah, though lots of money in play can pull some strings I guess.

[Beshr @beshr]

@thorstenball How much of the cost is Amp's and how much is Fable?

[Beshr @beshr]

I get the AI slop problem, but not accepting all public prs won’t solve it and feels like a dick move. A vouch system like Ghostty would’ve been a more understandable path.

[Beshr @beshr]

I think this is a mistake. Ladybird would’ve been an obscure niche hobby project if it wasn’t framed as an opportunity to contribute to a real new browser and a way to learn from Andreas, now that the project is on its feet, they decides to ditch their roots.

[Ladybird @ladybirdbrowser]

Ladybird is moving into a new phase as we work toward our first alpha release.

We are tightening how code enters the project: going forward, code changes will only be introduced by project maintainers, and we will no longer accept public pull requests.

https://t.co/iauF4r9f3q

[Beshr @beshr]

TIL, a Wayland client is not allowed to position its own toplevel windows, so GTK4 removed them https://t.co/AYvAXgt7ID

[Beshr @beshr]

This bitter anti-LLM sentiment is as insufferable as that "AGI will run the world" hype lunacy.

[DHH @dhh]

Empowering people to own and change their software was the open source slogan for decades. Now the grand democratization finally arrives, and it's all "yeah, but not like that" 🙄

[Beshr @beshr]

@pauliusztin_ @pydantic If you're into composable harnesses, check out Ossature, though calling it a "harness" might be underselling it, it's basically a full build system for spec-driven code gen ;)

Also built on @Pydantic AI under the hood (absolute banger of a lib btw) https://t.co/kVmIvwy8Sw

[Beshr @beshr]

@dbreunig I think it’s a sign of early optimization, or at least the wrong place to start? The more we anthropomorphize the less use cases we’ll find. LLMs we have now are quite powerful, we need better ideas for deterministic tooling and systems to build around them, to make most use.

[Beshr @beshr]

@DavidKPiano I think the point can be made whether you start with a prototype or spec. You do need to know what you’re doing/using to know how to write the spec well, but I don’t think you can feed that to an LLM. Code is one interpretation of a spec:

https://t.co/R5jbhsUdhQ

[Beshr @beshr]

SDD done right

[Beshr @beshr]

What Makes a Good LLM Spec: https://t.co/R5jbhsUdhQ

[Beshr @beshr]

@dok2001 @momito I’ve been working on Ossature for a few months and a lot of ideas are similar I think, but the corner stone for me is the spec to structure the intent https://t.co/ccOzFR70RT

[Beshr @beshr]

Maybe people who talk about this have always worked in perfect teams? My experience is different. People write crap code sometimes too. Besides, the spec captures intent, code captures one implementation of it.

[Ondřej Flídr @snipercze]

@CWood_sdf Yeah....this was 10 years ago, 2016, on https://t.co/wgRcXIFGvR

[Beshr @beshr]

Vibe coding is like Visual Basic. Everyone hates it, but in 20 years (after it inevitably dies) people will look back fondly at it.

[Beshr @beshr]

“That tension is roughly what specs become for AI-generated code. The spec has to be detailed enough that the agent doesn't have to invent the parts you actually care about, but it shouldn't be a thousand-line waterfall document trying to predict every branch”

[Ossature @OssatureDev]

AI-Generated Code Has No Author https://t.co/4wI8mp0qKb

[Beshr @beshr]

@dbreunig Some kind of a mix of the following: Start with a new context for every task when implementing a feature. Specs > Memory. Write behavioral contracts.

[Beshr @beshr]

@trq212 For software, the design of the harness will play a big role in how it helps engineers use it. My take on it is this https://t.co/ccOzFR70RT

[Beshr @beshr]

@trq212 Agreed. I think that the true benefit would be when prompting itself (or free-form prompting at least) is no longer needed by providing structure around how the user interacts with the model. Actually useful AI workflows will fade AI into the background.

[Beshr @beshr]

Writing a CHIP-8 Emulator from Spec https://t.co/kL0TTSRC63