Jaehoon Son

203 posts

Jaehoon Son banner
Jaehoon Son

Jaehoon Son

@knuceles

AI SWE @Tesla & AI Research @CMU Building Hermes. Find niche signals from your DB & turn them into personalized actions

San Francisco, CA Entrou em Ocak 2021
299 Seguindo114 Seguidores
Tweet fixado
Jaehoon Son
Jaehoon Son@knuceles·
Today I'm letting an AI agent start a real business. Completely on its own. No prompts. No hand-holding. I'm not touching it. Here's the setup: • $200/month Stripe wallet to buy whatever it needs • its own AgentMail inboxes • a full InsForge backend Then one rule: figure out the rest yourself. It does 99.99% of the work. I did the initial setup and stepped back. No human in the loop from here. It can write its own emails, spin up its own infra, spend its own money, and ship its own product. Honestly no idea what it builds. Or if it makes a single dollar. That's the whole experiment. Day 1 starts now. Following along in this thread 👇
Jaehoon Son tweet media
English
5
3
12
1.6K
Jaehoon Son
Jaehoon Son@knuceles·
The best thing to do for my business is not vibe checked. I let agents decide it for me based on data across Postgres, PostHog, and Stripe. At Hermes, your agents get context to user data you've been building for years that can rationalize and also run marketing on autopilot.
Jaehoon Son tweet media
English
1
0
1
28
Jaehoon Son
Jaehoon Son@knuceles·
There’s a $75 billion dollar market that serves enterprises to just answer the question “is this feature being used?” around 100 segments, 10 workflows, and 10 product/engineering guys. Today, there’s an AI system that can find product insights and autonomously act on it to win back the users. No matter the scale of the company, # of employees, or time you have. Only at tryhermes.dev
English
0
1
2
34
Jaehoon Son retweetou
Germain Hirwa
Germain Hirwa@GermainHirwa·
Today we're introducing the world's first Customer Operations Agent. Connect your product data and Hermes continuously watches customer behavior, finds churn risks, expansion opportunities, and onboarding friction, then automatically takes right action. Customer stuck onboarding → sends a personalized email Power user increasing usage → alerts sales Account showing churn signals → creates a success task Support issue blocking activation → opens a ticket Customer behavior → action. 👉 tryhermes.dev
English
0
1
3
107
Jaehoon Son
Jaehoon Son@knuceles·
At @tryhermeshq, we find signals directly from customer data sources, whether that’s revenue opportunities or churn risks, and take action like emailing users or creating a ticket. We started out with a dashboard that support all of our operation, but nearly third of our customers wanted to use this with their existing agents. That’s why we built MCPs and CLIs to integrate @tryhermeshq into any agents without the dashboard. The payoff? Speed at which our customers onboard themselves to the platform went from a day to minutes. A prompt with MCP/CLI can setup the entire @tryhermeshq within minutes as opposed to clicking/finding buttons in GUIs. Now more than half of our customers use the MCP/CLI feature the most and rarely the dashboard. The feature isn’t human doing the work, it’s the agents achieving a goal. For Hermes, that’s running the entire customer relationship for a business on autopilot with any agent. The interface we optimized for is the agents, not the humans.
English
0
0
2
23
Jonathan Wilke
Jonathan Wilke@jonathan_wilke·
I don't get the hype around CLI coding tools like Claude Code. Human-computer interaction evolved past the terminal 30 years ago for a reason. UIs won. Why are we regressing?
English
1.5K
59
2.7K
701.6K
Jaehoon Son
Jaehoon Son@knuceles·
SaaS companies still paying tools to send emails. sending is solved. it’s been solved for a decade. the thing nobody’s solved: knowing why to reach out to a specific user, right now. person A just hit their limit. person B has gone quiet for 9 days. person C invited 3 teammates last night. three different reasons. three different moments. and they’re hiding in your prod DB, not your email tool. Hermes finds the signals. the sending is an afterthought. tryhermes.dev/install
English
0
0
3
32
Kai Brokering
Kai Brokering@kai_brokering·
We made JARVIS for your computer. An operating system you control with your voice. The mouse and keyboard had a good run.
English
50
32
217
112K
Jaehoon Son retweetou
Germain Hirwa
Germain Hirwa@GermainHirwa·
Roast my startup: Turns your customer data into actions instead of dashboards Hey, I’m building Hermes. Not Hermes Agent, the OpenClaw alternative. A few teams are already using it, and I’m curious how other SaaS teams think about this problem. Most SaaS companies collect a ton of customer data. Product events, billing, CRM, support tickets, onboarding steps, all the usual mess. But a lot of it still turns into dashboards or reports that someone has to remember to check. A trial gets stuck during setup. An account suddenly starts using the product way more. A customer drops off after a support issue. A user comes back after weeks and starts looking at pricing. Those moments usually mean something. But someone still has to notice them, understand what changed, and decide what should happen next. That is what Hermes is trying to help with. Hermes watches customer behavior and turns important signals into recommended actions. Sometimes that means a customer email. Sometimes it means a sales task, a CS alert, a product ticket, a support follow-up, or doing nothing. The main difference from tools like Customer io, Braze, Hightouch, or BI tools is where it starts. Those tools are great once you already know the segment, campaign, dashboard, or sync you want. Hermes is trying to start one step earlier: “What changed in the customer base, why does it matter, and what should the team do about it?” Curious how other SaaS teams handle this today. When something important happens in your customer base, who actually catches it? Is it the founder checking dashboards, a growth person building rules, a CS person noticing manually, or do you already have something that works?
Germain Hirwa tweet media
English
4
1
6
394
Jaehoon Son
Jaehoon Son@knuceles·
Production DB to customer insights is smart. Using similar approach at @tryhermeshq, we find very niche signals hidden under the data sources and automously acts on them.
TBPN@tbpn

Y Combinator's @garrytan says he wants his new project GBrain to be the Postgres for agents: "The thing I realized is, a human can only keep 7, plus or minus 3, things in their head. But a computer with an LLM can keep about three Harry Potter books in its head." "Then, when you think about what most computer systems are, you should think of the Library of Alexandria — thousands, maybe millions, of books. It's even bigger than that. It's the whole internet." "You could basically take all the relevant info about customers, or any person that anyone at the company has ever even met. You can have that in like, 100,000 or a million markdown files that comprises everything that the business is. That's basically what GBrain can do." "The magic moment for GBrain is basically being able to take any 'book' that exists in your entire business, and making sure the 3 books that really matter for the thing you're trying to do are loaded." "And that's basically ASI. You don't have to write software anymore. You can just straight-up use Hermes agent or OpenClaw plus GBrain."

English
0
1
3
130
Hang Huang
Hang Huang@hanghuang_·
Excited for YC demo day :) Meeting with many investors today. Wish us luck!
Hang Huang tweet media
English
26
2
75
4K
Tony Chang
Tony Chang@tonychang430·
Insider: All the YC founders preparing for demo day
Tony Chang tweet media
English
3
0
29
1.5K
Jaehoon Son
Jaehoon Son@knuceles·
@luangm198 Thank you. I was already doing a lot of what digital businesses do inside my own Claude Code, so we’re pushing the limits of how far AI alone can go.
English
0
0
0
26
Luan GM
Luan GM@luangm198·
@knuceles This is a fascinating experiment! Giving an AI agent full autonomy to run a real business is next-level. The era of autonomous AI entrepreneurs is just beginning - can't wait to see the results! 🤖💼
English
1
0
1
16
Jaehoon Son
Jaehoon Son@knuceles·
Today I'm letting an AI agent start a real business. Completely on its own. No prompts. No hand-holding. I'm not touching it. Here's the setup: • $200/month Stripe wallet to buy whatever it needs • its own AgentMail inboxes • a full InsForge backend Then one rule: figure out the rest yourself. It does 99.99% of the work. I did the initial setup and stepped back. No human in the loop from here. It can write its own emails, spin up its own infra, spend its own money, and ship its own product. Honestly no idea what it builds. Or if it makes a single dollar. That's the whole experiment. Day 1 starts now. Following along in this thread 👇
Jaehoon Son tweet media
English
5
3
12
1.6K
Jaehoon Son
Jaehoon Son@knuceles·
It would need to find another way around! It’s already asking me for OAuth to Reddit, ProductHunt, and X to do marketing but I haven’t given anything. If it really wants accounts, it should find a way somewhere in the internet 😅. Though if it’s a real blockage like creating a legal entity or requiring identification, that would be one of the only times I step in.
English
0
0
0
42
OCD ReMaLf
OCD ReMaLf@OCD_ReMaLf·
@knuceles what happens when it hits a service that needs manual setup or identity verification? curious how you'd handle that without breaking the no-human rule
English
1
0
1
21
Dante K
Dante K@iliakakhadze·
Ilia Topuria boutta knock TF out of Justin
English
1
0
0
356
Jaehoon Son retweetou
Carson Lin
Carson Lin@lincarson_·
a customer shared one of their @tryhermeshq replies with us another dormant-user check-in turned into a thoughtful feature request. this is the kind of signal you miss when outreach is generic. 🫡🫡🫡🫡🫡🫡
Carson Lin tweet mediaCarson Lin tweet media
English
1
5
6
415