Cameron
@camhahu
Followers
262
Following
3K
Media
97
Statuses
1K
Husband | Founding engineer @enhance_labs | Ex-@Amazon | Blog: https://t.co/7TRiFcNo0V
Australia
Joined May 2024
I got really bothered by this so I built a CLI agents can use to control the browser no benchmarks, but feels faster than other things I've tried (MCPs, browser extensions, etc) install, run `browser add-skill opencode` (or claude, etc)
Tool calls are absolutely the wrong interface for LLMs to test things in the browser. Too slow. Integrated browsers/forks are also wrong. I understand the freedom but that makes them closed ecosystems. Complicated, can't hack on top of them, etc. We need something that can do
1
0
2
"I also get some programmers are eager to tune it out. The hype drones on, the fantastical claims are still far off, and there's uncertainty where this leaves the profession. But that's not reason to miss out on this incredible moment in human history!"
world.hey.com
At the end of last year, AI agents really came alive for me. Partly because the models got better, but more so because we gave them the tools to take their capacity beyond pure reasoning. Now coding...
54
74
773
MCP was too early. A CLI is a far better interface for an agent
the magic sauce of agents is just bash. you can fucking do anything with bash.
0
0
1
Seeing a lot of concepts for reviewing AI message threads alongside PRs - feels like building for the wrong future I'm not sure code review is for humans 12 months from now, in the majority case
0
0
0
Okay which one of my 21 unused domains am I going to throw ralph at I have a couple of ideas
0
0
0
the models are smart enough, your md files and prompting skills are not
20
6
153
I don't understand the long/verbose agents(.)md advice out there (but I want to) code is self documenting (provided you remove old code with something like knip) and good code progressively discloses itself, so why compress context into an agents file?
0
0
0
stop, please, just standardise https://t.co/CSsjM0huq1 and custom agents is all you need I have found skills to be useful so far but still dubious
0
0
0
If you get Claude to put aria labels on everything, your website becomes infinitely faster to navigate with something like https://t.co/ML9TpAwQJH, the CC chrome extension, playwright MCP, etc. Which makes verifying changes super fast Hopefully a second order effect of AI
github.com
The fastest, most token-efficient way for AI agents to control Chrome. - camhahu/browser
0
0
0
happy new year X thank you for giving me a timeline that is actually about things I'm interested in
0
0
1
can't seem to find a 'wrapped' too that shows my token usage accurately (since I've been switching between every AI coding tool this year). but this seems to be accurate from september onwards 2.9B tokens (btw gpt-5-codex is not my favourite model)
2
1
4
Does OpenCode just replace a whole category of code review AI SaaS now...?
0
0
0