Today is day 80 and the end of YC.
Final results:
- MRR +4100%
- Daily automation runs +650%
- Finally feel confident saying we’re making something people love
Legitimately sad
@ycombinator
is over but grateful it happened. So excited for what’s next.
Going for another 80
A YC batch-mate was quoted 16k from a consultant for a custom AI invoice processor
Their intern built it in 2 hours on Gumloop
GPT-4o now:
-Analyzes any receipt sent to their Gmail
-Extracts vendor + amount
-Categorizes it
-Adds it to their Notion DB
-Texts their finance admin
A YC batch mate wanted AI to convert their Youtube videos to SEO optimized blog posts
Took <1 hr to automate + add multi language translation
GPT4o instantly:
-Watches the video
-Writes a structured blog post
-Translates it to any # of languages
-Publishes them all to Ghost
Since YC ended, we've had 10+ demo calls a day
Every morning, Claude 3.5 Sonnet sends me detailed research reports about everyone I'm meeting
At 8am, I get a
-Text with a TLDR about my day
-Detailed email with research about every customer (ARR, company summary, industry etc)
A PR firm reached out wanting to have LLMs monitor news stories about their celebrity clients
It took ~20 minutes to fully automate
It watches their Google alerts, reads each new article, determines the urgency and writes a detailed daily report
Another YC founder told us they browse their product's subreddit weekly for feedback
We automated it. Claude 3 now analyzes top posts and emails them a full report
It spots trends in user complaints and areas they can improve. Generating a formatted report with sources cited
YC asks us every week what our most active users are doing
We got tired of digging through logs, so we automated daily usages reports with GPT
Using live metrics, it emails us a breakdown of days most active users, their most used integrations + free tier users to reach out to
We used to manually search LinkedIn for info about our new paying users
Now Claude does this for us
- It finds their personal LinkedIn
- Analyzes their complete work history
- Finds their company website + summarizes what they do
Instantly dumping this into our company slack
YC suggested we look for trends in our user feedback
~35% of our support requests last week wanted help with scraping complex websites
So we built an autonomous LLM web scraper
Describe your scraping task and have GPT independently click, scroll and type it's way to the data
YC told us 'it's sexy to get excited by the boring stuff ', so we built an autonomous PDF form filler into AgentHub
Upload any PDF + info and have AI go page by page filling out each field intelligently.
It's boring but possibly the most useful feature we've ever built 😅
We
YC said we needed to step up our outbound sales so we fully automated it.
Took ~1hr to automate scaled (non-corny) personalized LinkedIn outreach with AgentHub.
Loaded a sheet with 400 ideal customers, each LinkedIn profile is read (position, work history, recent posts) and a
YC suggested we learn more our new users so we fully automated background research + personalized emails to book an intro call.
Every new sign up triggers:
1. A Slack notif with background info about their company (ARR, industry, summary of what they do, country)
2. A
Another YC founder mentioned they browse Product Hunt daily for people using no-code website builders (their target audience)
We fully automated this for them with GPT
They now get a daily email with links to every launched product using Wix, Webflow, Framer, Squarespace etc
Our user research is 100% on auto-pilot
Our Gumloop flows do background research on every new user, dumping it in a massive Google Sheet for analysis
It finds their:
- company LinkedIn, scraping headcount, location and industry
- ARR with Apollo
- job title on their LinkedIn
Another YC startup with 4,500+ companies in their HubSpot wanted LLMs to help them keep track
This GPT4o automation:
-Waits for requests in Slack
-Summarizes all previous interactions
-Reads recent emails/notes
-Spots the 🔑 decision maker contextually
-Replies with a report
This live demo captures the exact moment I met my co-founder, 7 weeks before we got into YC.
My previous co-founder had quit out of the blue that morning. I was honestly sad as hell and went to this
@LangChain
meetup as a distraction.
The host asked if anyone else wanted to
We had over 200,000 nodes run on AgentHub yesterday alone🥳 A surprisingly large chunk of that comes from people building AI web crawlers
Here's one of my favourite new use cases
Thousands of real estate listings scraped and processed with LLMs directly into their Google sheets
One of our new users automated the most time-consuming part of their job by building an AI Slack agent
It monitors their company Slack for new lead msgs, instantly researching, categorizing, summarizing each one directly into their Notion db
took ~20 mins to build on AgentHub
@khyledhanani
You can add if-else statements within the flows + ask GPT to fallback on 'unknown' when it's unsure to gracefully handle failure scenarios.
I tried to keep this example flow simple for demo purposes but it's easy to tweak to handle any sorts of failures.
We built a simple tool that lets you prompt the 6 most popular LLMs at the same time
- Claude 3 Opus & Sonnet
- GPT4 & 3.5
- Gemini Pro
- Perplexity 70b
LINK:
I can always tell if I want to work with someone by how much energy I have after spending time with them
I dug up the first DM between my cofounder and I
It was the first time I'd met anyone in Vancouver as excited to build a startup as I was. It should always be this obvious
We’re looking for someone to join the AgentHub team as a founding engineer
If you:
- are legitimately excited to build
- love creating delightful UX
- are resourceful or an ex founder
Please apply, we’d love to meet you
DMs are open if you want to send work you’re proud of
When I decided to build a startup, I sent so many emails trying to validate ideas that I found a bug in Gmail which let me guess anyone's email address in a few seconds.
Here's how it worked (it's fixed now):
When you enter an email address into the recipient field, you'd see a
I did the math today.
@rbehal1729
and I have spent over 4500 dev hours building AgentHub.
~10 hrs per day x 6 days per week x 40 weeks x 2 ppl.
random messages from people enjoying what we built make it all feel worth it
A̶g̶e̶n̶t̶H̶u̶b̶ -> Gumloop
We've officially rebranded to 🎉
We've also released some fundamental platform updates that make building powerful AI automations even easier.
Better name, better product.
We automated our industry search on LinkedIn this morning.
Deploying internal tools like this in a few minutes is pretty satisfying.
Timed this build end-to-end on AgentHub, 22:07. Frontend is the AgentHub 'user view' and running a very simple automation.
Focusing on
@deedydas
great point. I think the key to doing this well is to extract info from existing quality content, not generate it from scratch with AI
I agree, pure AI generated content is almost always garbage
In cases like this, the video is dense with actual thoughtful opinion and this is
We launched on both Product Hunt and Hacker News this week.
Here's a breakdown of both launches and why 1 HN vote feels like 20+ PH votes value-wise.
(❗️Community noting myself at the end because we had an unfair advantage)
🧵 (1/n)
Day 23/80 progress update:
📈 2.5x MRR
📈 11x usage on AgentHub (automation runs/day)
❌ discovered that I need regular breaks and need 7+ hrs of sleep. Lost my sense of optimism without those.
My gf said this first tweet was dramatic (it was) but I literally moved to be closer
the power of no-code development 💪
we built a small but powerful web app with
@weweb_io
+
@AgentHub_AI
🤖🚀
analyze any subreddit (in 30 seconds) for trends, pain points, emerging trends, and startup ideas💡
try it here (free) 👉
I’m not qualified to give advice on much but I do know the YC interview process pretty well. I’ve done it 4 times (0.25 K/D).
Here’s what to expect and my top pieces of advice if you’re interested.
𝗥𝗲𝘀𝘂𝗹𝘁𝘀 𝗳𝗿𝗼𝗺 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗻𝗴 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻:
23% response rate. 70% of responders booked calls.
𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀:
Some intro calls are worth 100x others. Focus on that minority. Packing schedule with calls is not sustainable or fun.
Going to start throwing
YC said we needed to step up our outbound sales so we fully automated it.
Took ~1hr to automate scaled (non-corny) personalized LinkedIn outreach with AgentHub.
Loaded a sheet with 400 ideal customers, each LinkedIn profile is read (position, work history, recent posts) and a
@onionshorts_
They initially had a categorization step so Ai read every email which worked but switched to using a gmail inbox filter that labelled it as a receipt if it had certain key words + had an attachment.
That’s what they settled on because it’s more efficient.
@lazerwalker
To an engineer, anything is scriptable with a few APIs and some python. If you're not technical, building and hosting something like that is completely out of reach.
'Extremely easily' is a relative term.
That's exactly why we built Gumloop :)
Excited to share that AgentHub has joined the
@ycombinator
family.
We're on a mission to make AI powered automations cost-effective, reliable and intuitive for anyone to build.
Check us out on LaunchYC 🚀:
We met an alcohol distribution business that was manually placing 30k orders a year.
We used AgentHub to automate their entire workflow with LLMs.
Here’s a write up about how it works.
@garrytan
I whipped up an AgentHub (YC W24) automation that summarizes long youtube interviews and generates a single pdf with GPT's analysis of the interviewees character + the interview summaries.
I pasted in 3 of your interviews to test
(Highly recommend opening on desktop 🫣)
Rahul building entire features from the backseats of Waymos has become weirdly normal.
The guy literally never stops building. Most determined engineer I’ve ever worked with. By far.
@rbehal1729
from
@AgentHub_AI
has that dawg in him for sure
this is him on customer support on the car ride back from the pg talk. did the same on the way to the pg talk.
absolute legend
Function calling is by far the most underrated aspect of the GPT API.
I wrote a bit of a love letter to functions here and talk about how we use them at
@AgentHub_AI
What if you could build a
#GPT
𝗮𝗻𝗱 its tools with no-code.
Our new Agents feature lets you build assistants with access to powerful custom workflows.
Automate your task once, give it to your assistant and never perform it manually ever again.
I read my own accounts posts to check the news.
I built an autonomous Twitter bot with
#GPT4
to share the latest AI updates.
Here’s a short write up of how it works and how to make your own.
@geepytee
Surprisingly not cringe. Using a mixture of an example messages + words to avoid (like synergies, sync etc) works great.
I also categorize the profile into one of 3 customer categories as part of the automation. Each one has its own prompt with slight modifications. The prompt
Sales related automations are quickly becoming a user favourite on AgentHub.
This one generates personalized cold emails from your sales leads. I wrote a small blog post to break down how it works.
Use AI to generate thousands of ready-to-send personalized sales emails directly in your inbox 📨
Here's a blog post explaining how we do it with a simple AgentHub Automation:
OpenAi's functions feature is a gamechanger for applying LLMs to business use-cases.
No more begging GPT to return structured data.
We just added functions support on AgentHub and have seen the reliability of our pipelines skyrocket.
@gonza_nardini
Thanks. People are using AgentHub to power their own products.
Any automation can be triggered via api or embedded into your website for resale.
This specific build is calling our metrics api (with a call api node), passing all the data into GPT asking it to gen an HTML report
AgentHub has received some massive updates over the past 3 months🗓️
Watching old demos back helps put it into perspective
Here's the first demo we ever recorded of a twitter automation VS that same automation today
Today's demo (1/3)
We released a new automation framework on AgentHub
Here’s a video of it reading any article or website and tweeting about it for me.
We’re going to open source our automation library so we can build along side the larger community.
Overall:
I'd take a lukewarm HN post over
#1
on PH any day.
HN ~ speaking to a captive audience (who might kill you)
PH ~ putting your ad on a highway billboard with most drivers being people you already know or Waymos 🚙
@OfficialLoganK
@OpenAI
Our user automations on
@AgentHub_AI
are failing due to our org being throttled. We’re hitting the 10000 tokens/min limit 🥲
We applied for an increase Friday. Any chance you could review our application sooner 🙏
@rbehal1729
@santygegen
@ycombinator
+1
I also did the entire batch remote because of immigration issues.
Missing out on the in person stuff and grinding without other founders around was lonely. Still really enjoyed the process though
@OfficialLoganK
@OpenAI
✋
We help businesses automate any workflow with our node-code builder.
Here's a writeup of how some of our users send thousands of personalized cold emails with GPT.
Join me and the Vancouver AI community next week on Oct. 11 for an evening of learning and networking! 🤗
We'll be discussing building with AI and sharing demos. Want to showcase your own demo? You can! 🚀
Maybe meet your fellow co-founder! 👯♂️
Link in 🧵
GPT4 writes all of our public documentation 📜
This simple automation...
1. Reads our code 👁️
2. Generates detailed markdown docs for each operator 🧠
3. Raises a GitHub PR ✍️
@gonza_nardini
I was able to solve my problem in minutes instead of hours. Probably would never have had the time to build this otherwise.
I can also change it on the fly whenever I’d like.