Trying to make the web better! VP of Applied AI @Netlify by day Indie Hacker by night
ATL, USA
Joined August 2011
- Tweets4.3K
- Following2K
- Followers1.2K
- Likes6.9K
hi I helped build this. let's get claude connected to the world! ask me questions!
It’s now easier to build plugins for Claude.
We built a new portal to submit your plugin, track review, and see usage.
Plugins package MCP and skills, and are becoming the way to build for Claude. MCP usage across Claude products is up 110x this year!
claude.com/blog/build-plugin…
"Update the button text to say place order"
1. The transcription produces the instructions as text
2. Text parser recognizes "to say..." and extracts "Place order" as the replacement
3. Target resolution IDs the button using the pointer, it's name, or recently added item
4. Code updates the button's label and renders it in React (all front-end)
Never thought I would be into slack notifications remixes lol
Can’t decide if dancing to this at the club would be healing or traumatize me further
If you want to see an example on how having the very best AI model and harness doesn’t result in better/faster execution - in fact it slows you;
It took the Claude Code team 16 months LONGER (!!) to add support for AGENTS.md than OpenCode & Codex, 12 more vs Cursor
it had nothing to do with how hard it was and everything to do with controlling their own destiny in a way they think made the best strategic sense. It's true that not every agent works as well with the same context but beyond that, they want to do their own thing. Allowing support for both paths is definitely the way. I'll also say that, having been part of those conversations with agent builders in the early days of all of this trying to get these agents to standardize on _something_, zero of them wanted to be the first and they all had the same mindset. It's also worth noting that, the others were more incentivised to do it to support cross-compatibility so they can win over more folks. Claude code is still the dominant coding agent they didn't (and still don't) have that problem. Similar to how noone could get OpenAI to do anything driven by someone else. At some point, while it might be true that different context favors models differently, if there's no material proof over time then that justification falls over. So, rationale becoming weaker as the months roll on and tobi-level market pressure, it just makes no sense to hold back any more. But it was certainly a combination of things not simply technically hard or Tobi has the say
Experiencing FOMO while I wait for my invite to Jev (by @typesafeai).
We need the mailto: of MCP. Meaning, I want to click a button on a website and it adds the MCP to my agent.
Sean Roberts retweeted
we're on netlify!
TypeSafe's Jev answers typed questions about your program state instead of writing prose. You hand it the state and declare the answers you are willing to accept. A routing question can come back as sales, support, or spam and nothing else, which means you can branch on it directly without parsing anything.
netlify.com/changelog/typesa…
I definitely think Jev is a mind shift for what we’ve thought about with models, this is my breakdown for how to use it
Replying to @Netlify
Here is the full walkthrough if you want to watch it work.
A real support ticket goes in, and Jev comes back with the team that owns it, a frustration score, and a probability that it reads as urgent. The confidence value on each answer decides what gets handled automatically and what a person sees.
This video is larger than Cloudflare's 512 MB cache, so it can't be played through. More donations are needed to cover a larger cache. Donate
TypeSafe's Jev answers typed questions about your program state instead of writing prose. You hand it the state and declare the answers you are willing to accept. A routing question can come back as sales, support, or spam and nothing else, which means you can branch on it directly without parsing anything.
netlify.com/changelog/typesa…
insane @vercel_dev got access today and already shipped to production !!! congrats!
Jev from @typesafeai is on AI Gateway.
Build agents that decide, route, score, and stop in milliseconds:
𝚊𝚠𝚊𝚒𝚝 𝚎𝚟𝚊𝚕𝚞𝚊𝚝𝚎({
𝚖𝚘𝚍𝚎𝚕: '𝚝𝚢𝚙𝚎𝚜𝚊𝚏𝚎-𝚊𝚒/𝚓𝚎𝚟',
𝚜𝚝𝚊𝚝𝚎,
𝚚𝚞𝚎𝚜𝚝𝚒𝚘𝚗𝚜,
});
vercel.com/changelog/typesaf…
Nice! Evaluations is a good start for this. @justKDeng we should connect and bring it to @Netlify's AI gateway too and make it available for all of our builders to incorporate across the stack
Check your emails! We just let off a huge batch of people from the Jev waitlist. Getting as many people off waitlist today as possible.
I am here to predict the (partial) death of your favorite JavaScript frameworks and libraries…
In the age of agentic coding, I seriously doubt that frameworks that provide a lot of implicit magic that helped humans be productive will be the successful frameworks. IMO, code will need to read more explicitly. And frameworks, APIs, and libraries must evolve.
I tend to agree, though, while humans aren’t necessarily asking for frameworks or are having their own opinions as much anymore. AI has been trained on all of the prior opinions. So we found that framework decisions end up still being a very important driving factor but it’s the agents making them not the human agents aren’t deciding to do something basic and then building on that. They are deciding to jump all the way to a framework abstraction and I believe it’s whenever that becomes less the case than this future will actually happen
simple example someone gave me: the guardian model that most harnesses have implemented could use something like this
It will give you a confidence score for each available value from a finite set of values that answers your prompting. So imagine a prompt to generate a config based on a natural language prompt.. instead of getting AI telling you the exact configurations with the fields transformed to "myfield": "somevalue", it tells you its confidence of which value is right "myfield": {"value": "somevalue", "confidence":0.788} So if you know all possible values for some structure you can use jev to get the confidence score of how each value might meet the requirements of the prompt. The benefit being, if you have the ability to put a question and accept an answer that way, you can do so with much better speed and accuracy
🚀 Plugins are now live in Muse Code!
Install from a marketplace or local bundle, enable/disable in one step, update + remove with ease, and review runtime capabilities, all from /plugins
Extend Muse Code your way 🔌
Install: curl -fsSL dev.meta.ai/install.sh | bash
G2i has seen an explosion in revenue growth. The path they took to get there is... surprising. In this episode of The Build Log, founder @gabegreenberg met with host Jason Lengstorf to share the reasoning behind making some truly wild-sounding decisions at @g2i_ai
This video is larger than Cloudflare's 512 MB cache, so it can't be played through. More donations are needed to cover a larger cache. Donate
Gabe Greenberg's $100 million bet on the future of his company
G2i has seen an explosion in revenue growth. The path to get there is surprising.
Diogo Almeida@CompleteSkeptic
Sep 15United States
United StatesConnected via United States App StoreAccount-level information, not a live location or per-post device.
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
Diogo Almeida@CompleteSkeptic
Sep 15United States
United StatesConnected via United States App StoreAccount-level information, not a live location or per-post device.
The gains aren’t free: Jev can't generate text
Comparing Jev vs LLMs side-by-side makes the trade-off clear
Fun fact: replacing sequential computation with parallel is the same way Transformers leapfrogged RNNs
Amazing! We have ..lots.. of workloads and customers we’d love to put this in front of 10s of millions of devs on Netlify @CompleteSkeptic
Quinn Slack@sqs
Sep 12United States
United StatesConnected via United States App StoreAccount-level information, not a live location or per-post device.
Amp is now free to use when you bring your own compute and model subscriptions/keys.
No more limits or fees for BYOK.
ampcode.com/news/free-agent
Quinn Slack@sqs
Sep 13United States
United StatesConnected via United States App StoreAccount-level information, not a live location or per-post device.
Let's do it! I will get it started now. I just meant this to be a pricing announcement but I guess the early access to AI gateways dominated, lol
Cursor Origin 🤝 Netlify. That team is doing some of the coolest explorations in what new git flows could look like
Cursor Origin repos can now deploy on Netlify.
Push, preview, and ready to rollback in a click. Live in beta.
netlify.com/blog/cursor-orig…