He/Him Content Creator Core team, @solid_js Core team, @ripple_ts piped.video/@devagr twitch.tv/devagrawal099
Wichita, KS
- Tweets 19,488
- Following 1,629
- Followers 15,335
- Likes 65,527
Pinned Tweet
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.
3
2
50
15,902
tired: rewriting in rust
wired: solving real problems
The wildly opposite ai programming takes are a sign the folks calling it over are missing what's possible, because they were limited by what they came from. Dream bigger!
others: "it's over, guess it's all rust from here"
me: "omg there's so many things I can build in Elixir"
1
4
809
we need the effect-ts of frontend frameworks
and the rust of frontend frameworks
hmm i wonder how
20
1
77
14,167
i feel like i have lost most of my interest in LLMs and Agents
with system one models, the design space of AI applications just grew exponentially, and the majority of that design space does not use LLM Agents
it's crazy the more people haven't seem to realize this yet
11
1
59
9,439
the other factor could be i feel like the mainstream is still so far behind what LLMs on their own are capable of
harnesses are solved, but you won't be able to see the complete solution for another year or so
1
1
931
So this fucking thing will simply show up after 30 seconds of being on the page, occupy half the page, no way to dismiss it, the page is entirely unreadable and I can’t even scroll the page so it’s stuck
How does product engineering at a multi billion dollar company go so wrong?
We are truly doomed.
11
18
1,892
Dev Agrawal retweeted
gpuix now supports solid.js
build native apps with solid without any web views
gpuix.dev
30
54
624
39,200
i guess i'm gonna have to tweet atleast twice a day that it's not a decision model and LLMs make decisions too
Next version of OpenClaw uses a decision model to automatically decide between steer or queue.
(Lab feature, we support Jef* and API-compat (e.g. local models like Kef...) and ONNX variants)
1
1
567
If you're trying to run benchmarks on Jev the model you are measuring the wrong thing
The right way is to eval a system you have built using Jev
Break down the problem into lots of little questions and compose the final result out of it
If it doesn't give you the result you wanted, try breaking them down differently, and usually that's how you achieve gains
You don't want or need to change the model weights
1
521
So it’s just a coincidence that both OpenAI and Anthropic decided to release models at such low prices the week after Jev launches?
3
10
1,284
The one framework that can do it all
People said it wasn’t possible. Silver bullets don’t exist.
Probably not. But this gets PRETTY darn close
Everything from CRA to Nextjs to HTMX to Liveview
I've finally been able to put all frontend solutions on the same grid by recognizing them as a product of 2 axes and 3 core responsibilities. HTMX, LiveView, RSCs, islands, SPAs, Sync Engines...
None express what another can't. They differ in efficiency and ergonomics.
👇
3
1
26
5,620
Dev Agrawal retweeted
I've finally been able to put all frontend solutions on the same grid by recognizing them as a product of 2 axes and 3 core responsibilities. HTMX, LiveView, RSCs, islands, SPAs, Sync Engines...
None express what another can't. They differ in efficiency and ergonomics.
👇
20
23
219
33,828
> someone will (or has already?)
yes hi i have built and have been daily driving it for the past month
MCP vs CLI for LLMs is the wrong discussion. People are debating at the wrong layer.
Both work incredibly well as long as they are run through a repl like environment that can persist state. The funny thing about CLI is that CLI tools are usually accessed through BASH which happens to be a repl with persistent state (the file system), therefore cli works much better than mcp. But this isn't even close to being an intrinsic property of MCP. Just means that we need better harnesses.
Right now the best repl for this are:
- bash + fs
- jupyter kernels
- codemode type repls, usually quickjs
LLMs understand the concept of forward evolving a system to solve a need very well. This comes from the agentic RL. They understand how to change the state of a codebase or acquire data from databases and then operate on it similarly to humans. But without an execution environment, they actually cannot do this properly.
My bet? Sometime soon someone will (or has already?) create a embeddable, sqlite style mini execution environment that parses bash, typescript, or tool calls into a common IL execution plan that's easy to security check before executing. And design this specifically for durable execution environments. Then we will just connect cli, mcp, webmcp, whatever to that and it accepts any of the input modalities as they can all be represented as each other.
2
5
2,336
i'm trying to think of what a model would look like that sits somewhere between Jev and coding LLMs
maybe instead of a simple classification model (pick one out of these 5 options), it's a model that composes choices into a sequence on parallel
so input is some task plus some capabilities
and the output is something like
capability 1 -> capability 2 -> (capability 3 + capability 4) -> capability 5
it's a small and constrained enough space that it retains the "typesafe" aspect of Jev
at the same time it allows not just tool search but also tool composition to be done faster and cheaper without always having to resort to LLMs
3
8
997