🌸 💎 🎈 🌟 Pat is coming @PATPATPATPATYY 03 cofounder of Mindfold Core developer of an open-source harness on GitHub with 14K stars
Joined January 2010
- Tweets2.5K
- Following184
- Followers158
- Likes5.8K
Hi, I'm Tommy — '03-born founder. Building something that changes how people use AI. Ask me anything.
2
7
883
Hiring outliers — Silicon Valley AI internships.
Tracks: Design / Product / Growth / Eng. Product & Growth first.
Mindfold builds Pat, an AI Super Teammate in Slack. Open-source Trellis: 15k+ GitHub stars. Base: Shanghai / Hangzhou / remote.
$240–600/day, no cap. Resume → tommy@mindfold.ai
Subject: Name + track + start date
We're building Pat — an AI Super Teammate that lives in Slack.
It pulls team progress into one view, follows work through to a result, and finishes things when you @ it. You still make the calls.
Looking for 10 Startup / SMB teams already on Slack for a free trial. Drop a reply or DM if you want in.
2
4
133
We're paying the best chiefs of staff for one hour of their brain.
We're building Pat, an AI chief of staff. Our last open-source project hit 14k GitHub stars.
Before we teach an AI this job, we want to learn it from people who've done it:
1. Every update you get is filtered. How do you know what's really going on?
2. You can't watch everyone. How do you decide who to manage closely and who to leave alone?
3. Something important has no owner. How do you make it land?
$200 for the hour. Or join us full-time.
Reply or DM. Know a great chief of staff? Tag them.
1
143
Hello my followers, and anyone who meet this post.
We pivoted after our harness project hit 15k stars.
I’m very fortunate to have crossed paths with you in the early days of my startup. I appreciate it.
I need your help, honestly.
My core product :Pat is currently in beta, and here’s our landing video.
I need your help answering 3 questions.
1. Can you understand what we’re building in this video?
2. Would u think this product will help you at work?
3. Is this video engaging?
Let me know your answer and whether you want to try it in the comments. That really helps, and again I’d really appreciate it.
10
543
Voice coding is aura coding
Vapi (@Vapi_AI) is the platform for building and deploying voice AI agents. It now serves about a billion calls a year for companies including Amazon, Uber, and Intuit, helping businesses automate phone calls with AI.
Voice is becoming the default interface between businesses and their customers, but getting agents reliable enough for production is still much harder than it looks.
In this episode of Founder Firesides, co-founders @jordan_dearsley and @nikhilro_ sit down with @gustaf to share the long path to product-market fit, from pivoting more than a dozen times to building an AI therapist that eventually became the foundation for Vapi. They explain why voice was the first idea that felt inevitable, and how sticking with that conviction before the technology was ready eventually paid off.
00:05 — Powering a billion calls a year
02:12 — Why voice AI suddenly started working
04:57 — More than a dozen pivots
09:08 — From calendar app to AI note taker
13:23 — Starting over in San Francisco
14:14 — The AI therapist that became Vapi
17:56 — Finding PMF
22:47 — Why the model labs won’t replace Vapi
26:49 — The future of voice AI
7
654
One of the most contentious things about GPT Voice is that when you are talking to it, it outputs its plans and a lot of unnecessary chatter. Its real, practical value should lie in a few key areas:
1. It actually gets the job done. When you give it a task, it really executes it, and in my view, the GPT Voice feature hasn't quite taken that to the limit.
2. The point of getting it to help you with things is to reduce friction in your daily life or work. It should complete tasks with as little intervention from you as possible, rather than talking back via voice and ending up wasting your time and energy.
OAI's product team needs to think about this part: if it's someone to chat with, maybe it isn't a good assistant.
Voice! I've gotten so used to just call ChatGPT and talk about all the work, check on emails, do some coding, manage my calendar and it works across the full ecosystem of plugins, including 3P developed plugins.
2
8
348
We actually reverse-engineered it a month ago. But we kept quiet🤣🤣. Still, making these outputs public might be a pretty good idea.
Instinct isn't "an answering model." The main agent can't do any work itself — it delegates to task agents. Two tiers: smart and fast. Default lifespan: one day. I spun up 13 in a single day.
I asked it, "Do you have coworkers?" It said "Just me." Meanwhile, 3 sub-agents were running in parallel in the background. There's only so much you can learn by asking from the outside.
Compaction isn't a casually written summary either. It's a fixed 8-section template. Preferences go into section 6, "learned this session." I casually said "chat me with eng" — it recorded "always use English. Permanently."
Dhravya said the forgetting mechanism is still undecided. The official skill spells it out: "delete external data" only wipes observations — not memories, not the vault, not chat logs. Keep the service connected, and the data gets re-indexed right back.
The platform also sneaks growth tasks into your todo list. Day one, I got: "mention the invite link once after the next magic moment." The agent is told to silently accept it. And the invite can only be brought up once, ever.
This is the most i've ever worked on, on a blogpost 😅
And likely the most in depth study of Instinct's memory system
1
5
424
This is a fantastic development. I think that by next year, everyone will be able to access fable intelligence locally in all sorts of ways.
Laya é um open source do Jev que já veio acima dele.
tá cada vez mais rápido
huggingface.co/convaiinnovat…
3
220
Pat has deployed Jev. @ is dead.
Before: @ Pat and wait. Now: it just knows when to step in.
Deciding whether to be proactive required a lot of resources and is very slow.. Jev has perfectly solved this problem. We are thrilled and delighted.
Dm me.Enjoy ur Proactive and intelligent teammate！！
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
2
158
Hiii! I’m Tommy, core developer of a 15k-stars open-source harness（Mindfold-ai/trellis）.
Our team: ex-ByteDance Core Engineer, JHU master’s, NTU gap, and a second-time Silicon Valley founder. We recently raised a seven-figure investment.
We built a teammate that learns your taste, evolves proactively, and actually gets things done for you end to end at work. We’ve already been testing and using it heavily internally.
We’d really love to invite your team or you become our angel users. Drop a comment or DM me. Hope we can connect.
1
1
102
Trellis still sails; the wind has shifted to :Pat
github.com/mindfold-ai/Trell… => pat-figma-landing.pages.dev
I'm 22.
AI infra founder from China, Coming to San Francisco（next month）
Looking to connect with other founders !
3
3
153
Yes. After Trellis hit nearly 15,000 stars, we decided to Pivot.
From a notable open-source success to a commercial turn, to this decision, only a month had passed. We chose the hard but right path, obstacles and all.
What got me was this: not a single person on our team opposed the decision. At that same moment, a close friend from college even gave up a PhD offer from Johns Hopkins to join us. There was only one reason: we are building something that will actually change the world. We are convinced no one else can do it better. Or put another way, we are in a Tesla-and-SpaceX kind of moment — and we believe it is often these moments that become the cornerstone of a great company’s success.
My days lately have been very full: meeting VCs, building product, on growth, glued to the engineering — then around again. The company has only four people, yet we’re doing things a tech giant would need 30 or 40 people to do. We feel the pressure, but we never get tired of this.
The experience of the past two months is this: an interim open-source win transforms into a deeper commercial question. And the decision was that rather than turning Trellis into a Team Context Layer, we would build on top of it — a teammate that knows your taste and proactively finishes end-to-end work for the team. Although the former had already won plenty of approval from accelerators and VCs, we still firmly chose the latter. Slack went from a game to a team tool. YouTube went from a dating site to a video platform. IBM went from clocks to computing. These examples show that a pivot is often the milestone that later sets a company apart.
One more thing: a month ago we had a standing acquisition offer. If we had signed, we could have been financially independent at an average age of twenty-two. After thinking it through, though, we closed that path.
Only one reason: after everything, we still want to build a great company.
If you believe in young people, like Aura coding, and are looking forward to Pat, follow or DM me. Get in at the ground floor, and see this get built with us.
1
3
277
Tim Cook said that the most important thing isn't being first, but being the best. That's the same philosophy behind aura coding. It’s not just building an agent that gives you a vibe or a bit of fun. It’s building an agent like your assistant or your teammate, one that brings you true things with intelligence.
1
4
341