Dev Agrawal

AI isn't going to "eat up" anything. SaaS App stores Frameworks Higher level abstractions Frontend Product design Nothing is going away. Everything is changing. But the fundamentals are more important than ever. Stop with the fear mongering. It just makes you look dumb.

@sysstary7556 You have beginner energy and you just proved it as well

@RyanCarniato @jarredsumner get this man a max subscription already

Fable has made me distrust every other model I've used before. It isn't perfect, but when I have a nagging suspicion something won't work completely Fable identifies it without prompting. Every other model while thorough spews complete confidence even when pushed. A while back @theo asked for problems that certain models couldn't solve. I was working on some hard problems but nothing I threw at them they couldn't solve if spec'd well enough(ie tests). In some cases it wouldn't be as good as I could do manually but it was respectable. But when creating something completely new you don't know what you don't know. You work with the agent to spec it. You test hypotheses. You benchmark structures. You look for gaps. I let frontier models talk me into a place of comfort, even though I had concerns. In the past week Fable confirmed my instincts and drastically tightened up the work I had been doing. Poking huge holes in architectures that I'd been repeatedly told were streamlined, stable, hardened even why I posed the same questions. Having the back and forth swapping between it and GPT 5.5/Opus 4.8 to save credits for different tasks has made the difference even more evident. Even on simpler tasks like PR merges other models leave so much on the table. It reminds me of when I first started.. going from GPT 5.0 to Opus 4.5. Maybe even more so. The difference is so profound it makes it really difficult to use anything else at the moment. If I've learned anything in the last while I haven't been ambitious enough. The problems that I know have solutions but have been too hard to do are on the table. I fear financial pressures are going to eventually be the barrier here. It's too easy to make a tool like this a crutch when it is the only seemingly feasible way to close the gap to what has been thought previously not possible.

Never thought I’d see a Musk Theo crossover again after the first one Crazy full circle moment

@elonmusk @Ph4seOn3 It’s almost like he knows what he’s talking about

Good review of Grok 4.5

@zeeg @KentonVarda The real answer is - the harness should commit the changes made in a turn using the prompt you gave it as the commit message along with a concise summary of the reasoning.

@KentonVarda im the entire opposite: never writing a commit/PR description again build a skill or similar abstraction to make it as repeatable/useful as possible

I just declared a moratorium against AI-written change descriptions (e.g. PR and commit messages, also issues/tickets) from my team. AI was writing change descriptions that were worse than useless to me as I tried to review PRs: outlining details of the code that could easily be seen by looking at the code, but omitting the higher-level framing needed to understand broadly what the code is doing. I think people like having AI write these things because the output looks structured and thorough, which makes it feel professional in a way. But this isn't actually valuable. Concise, high-level descriptions are better for everyone. If I need to use my own AI to interpret what your AI wrote then something is wrong. Let AI write code, sure, but for the description, I'd rather see your prompt than your output. We could maybe have extended agents.md with guidelines on writing descriptions, but this seemed a bit pointless since a good, concise change description only takes a few minutes to write -- not a significant time savings to delegate to AI. At least, it doesn't take long if you understand the code -- and if you don't understand the code, then I'm definitely not merging it.

@theo What would a better benchmark look like?

Agreed.

@MatijaSosic A benchmark for long horizon coding, planning, and architecture,

@devagrawal09 What would you like to see?

Sounds like we need a new better coding benchmark for AI who’s working on one?

Rate limit reached, time to touch grass and eat sunlight I guess

@nexxeln Sounds like the company is finally gonna make some money

its over, we moved to slack

The stream is back this Friday! It's been quite a journey but I'm very excited to talk through where the Async Solid 2.0 features have landed. youtube.com/watch?v=_914md…

@0xkanishka Another part of it is having multiple agents implement the same slice using different models and picking the best one, which likely requires either worktrees or some alternative, and I didn’t wanna mess with that

@0xkanishka True. It’s one of the reasons why specter is still at level 6. There’s a lot of work that goes into building the coordination layer. It seemed non trivial at first but as I started building it I realized it’s a lot more complicated than I thought and moved it out of scope.

NEW VIDEO From Vibe Coding to SpecOps, 10 levels of Spec Driven Development for upgrading your AI coding workflows What level of SDD are you on?

@0xkanishka > It has no answer for running several agents in parallel Look at level 7 again lol, that’s literally the first thing I said, have multiple agents build slices in parallel

Solid breakdown. The rung that's missing: this ladder is all one builder → one app. It has no answer for running several agents in parallel - who claims a slice, who reviews, who merges, how dependent slices stay out of each other's way. That coordination is orthogonal to all 10 levels and it's where the real difficulty is. (It's what I'm building with Spec0.dev) Living around 3/6/7 myself, and skipping 10 on purpose - real inspectable git beats specs-as-source for me. Would love to hear your thoughts on Spec0!

Wish there was more @solid_js articles. Glad people are giving it a shot morello.dev/blog/solidjs-2…

@johncrickett @theMissionBoy “The most expert horse riders can outpace people sitting in a car but don’t know how to drive it or keep bumping into shit”

@theMissionBoy AI is not mandatory for coding at the highest level. It's very useful, but there are still plenty of talented people who can out-deliver many engineers who are using AI.

Why is this a problem? Isn't the goal to solve user problems. Does it matter if an engineer does that with or without AI if they can deliver more value than it costs to employ them?