Verifying humanity (in stealth) | Prev. early team @ MultiOn (first computer use agent) - evals + leading QA/customer support Don't outsource your logic.
Seattle, WA
Joined August 2024
- Tweets1.6K
- Following2.3K
- Followers221
- Likes1.6K
Pinned Tweet
I’ve spent the last 2 years obsessing over one idea and just shutdown my startup that was generating 6-figures a year to pursue it full time.
I thought I wanted to work on something sexy and flashy next, but it’s like I’m being drug into this by my ankles. This company needs to exist and I can’t convince myself to work on anything else.
[REDACTED] is a developing offline human verification for close personal networks like your family, friends or coworkers.
Our long-term strategy is to develop profitable consumer products that fund continued R&D into high-impact, open source solutions; even when those solutions aren't independently profitable.
In doing so, we aim to establish ourselves as an industry authority on simple, practical, accessible defenses against AI-enabled threats, making harmful activities increasingly difficult and economically unsustainable at scale.
It's day three and there's a certain peace in knowing exactly what you need to do, even if you don't know exactly the path you'll take to get there.
Follow along for the journey. We're looking for capital.
When wetware becomes mainstream we're going to start hearing people say things like "so you'll eat an animal but you won't run compute on it?"
today, we introduce a little mischief to the world.
for four years, we’ve been brewing real brain cells on electrodes to understand how they compute. now, we’ve turned those discoveries into a video model anyone can use.
from 0.5¢ a second, mischief 1 is at least 5x faster than its base and the cheapest image-to-video model on @fal and @OpenRouter. soon to be on @awscloud marketplace.
it’s not a delusion anymore. biocompute is real, commercial and this is just the beginning.
happy halloween.
> The QA and support seat usually sees those patterns long before the model team does.
This is very true. Leading QA and customer support at a SOTA AI lab was genuinely a curse of apollo simulator.
Real users are where computer-use agents meet every odd login flow and pop-up a demo never shows. Recording each failed session as a replayable trace, with the page state at every step, tends to be what turns support tickets into a regression suite. The QA and support seat usually sees those patterns long before the model team does.
Who are the best impact driven angels and VCs for something like this?
I’ve spent the last 2 years obsessing over one idea and just shutdown my startup that was generating 6-figures a year to pursue it full time.
I thought I wanted to work on something sexy and flashy next, but it’s like I’m being drug into this by my ankles. This company needs to exist and I can’t convince myself to work on anything else.
[REDACTED] is a developing offline human verification for close personal networks like your family, friends or coworkers.
Our long-term strategy is to develop profitable consumer products that fund continued R&D into high-impact, open source solutions; even when those solutions aren't independently profitable.
In doing so, we aim to establish ourselves as an industry authority on simple, practical, accessible defenses against AI-enabled threats, making harmful activities increasingly difficult and economically unsustainable at scale.
It's day three and there's a certain peace in knowing exactly what you need to do, even if you don't know exactly the path you'll take to get there.
Follow along for the journey. We're looking for capital.
If you’re building a browser or a computer use agent, you need to pay attention to this.
I was part of the early team at MultiOn, where we built the worlds first computer use agent to go beyond a flashy demo, deploying to 30k real users.
As head of QA I was consistently researching and testing improvements to reliability, from replacing our orchestrator model with a VLM, to deploying sub agents that only interact with specific UI elements but we could never get speed and reliability to be where it needed to be. Then as models improved, surprisingly, MultiOn’s agent swarms actually got slower and less reliable.
After the first time that I talked to @Andrey__HQ and tried @_agentweb, I instantly realized where the entire industry went wrong - AgentWeb is elegantly solving the same issues that kept me up at night and have been holding computer use back from being genuinely useful.
This is a full circle moment for me and it’s amazing to see.
Introducing AgentWeb: reliable web infrastructure for generalized agents.
47/47 real world tasks on live sites. 1.9s median. $3 total.
Request access: agentweb.us
Julian Brooks retweeted
at some point we’ll just accept that the alien-ghosts can use a lot of mechanisms that we use
Today on the blog, learn about our new instance of a “sleep” phase in continual learning, which enables LLMs to continuously integrate new facts and enhance reasoning while mitigating catastrophic forgetting or static knowledge cutoffs.
More →goo.gle/4iICkyD
ALT Overview of our sleep framework: Input streams from the active phase are stored in short-term memory before entering the sleep cycle, which bifurcates into Consolidation (transferring fragile short-term tokens into expanded long-term parametric memory using low-rank structures and MoE) and Dreaming (autonomous self-improvement via synthetic routing with random experts and RL refinement).
If you need a quick QA check on your app just DM me!
I've previously lead QA at MultiOn, the worlds first computer use agent. Now I'm starting something new and need to bootstrap to get some early traction.
If you've been pushing to prod non-stop or have a big demo coming up soon, dont panic; I'm offering a 48-hour QA sprint that'll keep you moving fast with confidence. If you know anyone like this, I'm paying out 25% comissions on any closed deals.
First 30 minutes is free, just point me to your app, software, website, etc. I'll do a quick pass, and let you know what I find.
Money back garuntee if for any reason your not happy.
I've previously lead QA at MultiOn, the worlds first computer use agent. Now I'm starting something new and need to bootstrap to get some early traction.
If you've been pushing to prod non-stop or have a big demo coming up soon, dont panic; I'm offering a 48-hour QA sprint that'll keep you moving fast with confidence. If you know anyone like this, I'm paying out 25% comissions on any closed deals.
First 30 minutes is free, just point me to your app, software, website, etc. I'll do a quick pass, and let you know what I find.
Money back garuntee if for any reason your not happy.
Combine consumer brain computer interfaces and the next generation of kids is actually cooked
I want a platform where philanthropists and sponsors can give money or credits to researchers and builders who want to test weird ideas and create cool things, without worrying about a company. Just provide a cost‑of‑living stipend, unlimited tokens, and watch what happens.
Talked to someone who just got out of federal prision recently.
He was charged with insider trading around the time that GPT-3 was released. He didn't even see the jump from 3-3.5
Once he was done excitedly showing me it creating workout plans and answering all of his questions about things he missed, I sat him down with GPT-6 Astra Ultra and told him to pretend hes talking to an alien that can do anything. For the first 30 minutes he looked like a kid again. For the last 30 minutes hes been having an existential crisis.
I’m looking for new opportunities! Previously led QA & CX at MultiOn (the world’s first computer-use agent) and helped build agent evaluations at its spinoff, AGI Inc.
Overall, I’m a massive generalist and professional coal shoveler 💩
DMs and intros appreciated!
Julian Brooks retweeted
Unfortunately, with access to an abliterated DeepSeek V4.1 model, there is little value to KYC holding back the frontier models now for cyber.
In fact, making it hard to secure systems w extra friction in American Enterprise is hurtful to national security.
Julian Brooks retweeted
If you’re building a agent
Always go the harness route first. Do not build a complicated 20 step workflow using a bunch of if else statements or langgraph
Use a harness as your agent eg. Claude code, codex, deep agents, glean etc
You’re just shooting urself in the foot off the bat
Trust me, prompting works surprisingly well