2x Founder | Building https://teapaw.org/t.co/WF2tTNaMgy
Toronto, Canada
Joined April 2010
- Tweets415
- Following426
- Followers123
- Likes147
Eric Kim retweeted
Breaking: Browser Use + Jev = Ultrafast ⚡
Findings flights took 7s and cost only $0.0039 🤯
> new action space every step
> DOM state space
> small LLM fallback to type
(this video is at 1x speed btw)
Built a tiny open source browser agent. try it below ↓
@typesafeai is improving my preferably deterministic workflow requests by 5-15x in speed and apparently 100x cheaper than the cheap LLMs.
Imagine now in your workflows:
0. your agent repeatedly makes the same decision for inputs
1. agents PoC structured output with LLMs
2. agents create evals
3. agents prompt for @typesafeai style requests
4. requests in series, and agent can do task 1000x cheaper.
Claude Code usage has become Cursor.
This is what happened.
1. VS Code + Copilot
2. Cursor
3. Claude Code
4. Codex
5. Kimi?
$200 USD for a few queries. "That's the cost of LLMs!"
Well, probably not worth it in this capacity.
Ran Spar.so on hedleyandbennett.com today and found their og:image tag set to the literal string "[object Object]" on every product page. The link preview breaks with no photo. AI audit ftw.
One of the best parts of dogfooding your own product is that it forces honesty.
You stop thinking about the product as the builder and start experiencing it as the user.
The bugs feel more annoying and the wins feel more real. Try out Spar.so on your site today.
How are you supposed to select a part of the message to copy on iPhones? Do I always need to copy the entire message and paste it elsewhere so I can select a code or something?
This but with Flash MoE to run Qwen3.5-397B-A17B as an agent for long running deeper processing.
So cool!!
I'm running @nvidia's NemoClaw with Qwen3.5-27B entirely locally via Telegram. No API costs, no data being sent to anyone
NemoClaw is a security sandbox built on top of @openclaw that lets you restrict the files and networks your lobster can access.
I'm running the 4-bit GGUF via llama cpp on a DGX Spark. It's running pretty smoothly at 9 tokens per second!
The future might be local :)
LLMs training on our continuous input is essentially the human hive mind as it evolves. We’ve started to incorporate AI outputs as part of the seed in our own creativity. AI is actually using OUR brains to refine the collective consciousness. Not the other way around.
Towards Autonomous Optimization: the website that evolves.
The next generation of websites should identify their own weaknesses, generate hypotheses, test solutions, and implement winners. Continuously. Autonomously.
Early store deployments show 20%+ conversion lifts within 1-2 tests. Not joking. Impact is weighted across multiple KPIs and prioritized per site/industry.
Someone has to talk about this phenomenon where our our culture is being shaped by a few models. Our writing style is imitating the shitty LLM artifacts that we hate.
What the fuck is a "kicker"?
FUCK shadow dom for component libraries. Shadow dom should be drag and drop widgets with configurations that must not be overridden. Please think twice before you implement everything with shadow dom.