
Interviewing great AI founders. Presented by @fondocom
San Francisco, CA
Joined October 2024
- Tweets4.1K
- Following2K
- Followers4.4K
- Likes6.7K
BREAKING 🚨 Hypersonica raises €100M Series B co-led by @lightspeedvp and @generalcatalyst, with Plural, 201 Ventures and SPRIND, to accelerate development of its deep-strike systems, increase testing frequency and expand across Europe
hypersonica.com/en/news/hype…
TRENDING 📈 @Akashi203 says GPU providers went from charging $7.9/hr for B300s last month to begging to rent them at $4.6/hr with no commitment: “something big is happening soon”
START retweeted
The @YCombinator deadline is today.
I still remember when @daltonc called me at 9pm on Halloween.
I really thought I bombed the interview.
But no, he had great news. We got into YC.
W27 apps close tonight at 8pm PT.
If you're on the fence, do it.
BREAKING 🚨 Northwood raises $375M Series C co-led by @a16z and DFJ Growth, with Alpine Space Ventures, Stepstone Activate, In-Q-Tel and others, to accelerate delivery on existing contracts, hire talent and build out its global ground network
prnewswire.com/news-releases…
TRENDING 📈 @karpathy built HackerTube over a weekend: Claude filtered 43M Hacker News comments from 20 years down to 6,454 videos people keep recommending, in about 2 hours with Opus 5.5
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
START retweeted
those costumes tho 👻
@skygodkingdom @JoshConstine @kmkaric
@Fondocom x @lovart_ai x @hellyeah_ai x @ManusAI x Press Club x Paradigm x @neobrowser
START retweeted
Everyone wants to invest once it's obvious. We are getting a ton of inbounds after the recent competitive announcements.
Still remember David, @Fondocom giving us this exact vision a year ago. Was great to meet and catch up with him at the recent YC alumni demo day! @avinashjoshi
BREAKING 🚨 @mintive raises a $7M seed round led by @cowboyvc, with Operator Collective and Burst Capital, to build infrastructure for the agentic enterprise
hubs.ly/Q04CW2C50
TRENDING 📈 @moinnadeem introduces Phonic-1 from @phonic_co, “the most advanced speech-to-speech model for agentic tasks,” and a $27.5M Series A led by @firstround with @Lux_Capital and @felicis
Introducing Phonic-1, the most advanced speech-to-speech model for agentic tasks — from @phonic_co, the speech-to-speech model company.
We’ve had a breakout year and raised a $27.5M Series A led by @firstround with participation from @Lux_Capital, @felicis, and Definition.
START retweeted
met an amazing YC founder yesterday.
@davj. founder of fondo. YC W18. doing well above $10M ARR.
helping startups with bookkeeping, taxes, and tax credits.
got me thinking about how i’d find customers for a business like his.
because almost every startup needs what he sells.
so where do you start?
i’d look for founders interacting with stories about problems they want to avoid.
> a founder sharing how they missed a tax deadline.
> another explaining how messy books delayed their fundraise.
> someone breaking down an expensive mistake they made with startup taxes.
then look at who’s interacting.
> the founders commenting.
> the founders asking questions.
> the founders reacting.
and those conversations happen under more than founder stories.
> an accountant posting about an upcoming tax deadline.
> a lawyer explaining what founders need to file after incorporating.
> a competitor sharing a customer story or answering a bookkeeping question.
i’d look at their engagement too.
because someone engaging with a competitor’s content could be researching solutions, comparing providers, or already using one.
worth checking. worth starting a conversation.
then i’d run the same play on content from fondo’s founder and team.
> post stories founders recognize.
> the deadline someone nearly missed.
> the bookkeeping mistake that took days to untangle.
> the question that keeps coming up on customer calls.
then identify the founders who interact.
the content helps you find people and the outreach starts a broader conversation about their business.
my message would be:
“hey [name], how are you handling accounting at [company] today? any questions we could help with?”
or, when there’s a relevant customer connection:
“hey [name], we work with [customer A] and [customer B]. you might know them. anything on the accounting side we could help your team with?”
use actual customers they’re likely to recognize.
- same stage.
- same industry.
- same founder community.
give them a reason to trust you and an easy way to reply.
if they say “we’re happy with our accountant,” move on.
if they say “actually, we’re trying to figure out our tax filings,” there’s a conversation worth having.
if they’re already using a competitor, ask how it’s going.
- maybe they’re happy.
- maybe they’ve been waiting weeks for an answer.
let them tell you.
here’s the playbook:
step 1: find founder stories about tax and bookkeeping problems.
step 2: find lawyers and accountants posting about relevant deadlines and obligations.
step 3: look at who engages with competing providers and their teams.
step 4: use @GojiBerryAI to surface those people and qualify them against your customer profile.
step 5: reach out with a useful offer to help. bring in recognizable customers when relevant.
step 6: publish similar content through your own founder and team, then follow up with qualified people who interact.
if almost everyone in your market needs what you sell, useful content gives them a reason to show up.
> your content.
> an expert’s content.
> a competitor’s content.
find the right people in those conversations.
ask how you can help.
go close some deals!
BREAKING 🚨 @share_willow raises an $8.5M seed round backed by Lerer Hippeau, Circadian Ventures, Twelve Below and others for its performance pay platform and AI agents for the trades
hubs.ly/Q04CW1L90
TRENDING 📈 @simonw shares a case study of a 10-person startup that ships 77,000 changes a month with AI agents: “my eyebrows raised at least a dozen times while reading this”
BREAKING 🚨 Neuron Industries raises an oversubscribed $10M seed round co-led by @FundersClub and @obviousvc, with @ycombinator, to build AI-native industrial controllers
hubs.ly/Q04CW0MG0
TRENDING 📈 @adcock_brett says @hark_labs launched Hark on Mac: “The same Hark you know and love, right on your desktop”
Today we launched Hark on Mac
The same Hark you know and love, right on your desktop
Download Hark for macOS: hark.com/mac
BREAKING 🚨 NanoForge AI raises a KRW 14.6B (~$10.9M) seed round led by Mirae Asset Venture Investment to build electronic-structure prediction and a robotic lab for new materials
hubs.ly/Q04CW0gr0
TRENDING 📈 @seanlie breaks down how @cerebras deploys multi-trillion parameter models: fitting the full model into on-chip memory, mapping it across multiple chips and handling many concurrent requests
How we deploy multi-trillion parameter models at scale
Today, let's break down the math for how @cerebras:
1. Fits the full model into on-chip memory
2. Maps the model across multiple chips
3. Handles a high # of concurrent requests
BREAKING 🚨 Dream emerges from stealth with a $17M seed round backed by @a16z, @svangel, @xyz_vc, and @ziggcap to build AI that automates the design and engineering of buildings
hubs.ly/Q04CV_jQ0
TRENDING 📈 @gmkurtzer says Ginux, a kernel built with AI in 11 days that boots Rocky Linux 10, was an experiment, not a Linux rival: “The development process was the experiment. The kernel was the evidence.”
People are missing the point of Ginux.
I didn’t create another kernel to compete with Linux. That is idiotic!
In fact I consider the code disposable. Think about that…
Ginux wasn’t the product. The development process was the experiment. The kernel was the evidence.
Could AI, directed by an experienced engineer, could build and validate an extraordinarily complex software system.
In just 11 days, I produced an independent kernel capable of booting Rocky Linux 10 with systemd.
That was the experiment. And while Ginux isn’t a complete Linux aligned kernel, the experiment is complete.
The research paper documents the methodology, results, and limitations.
Since publishing, people have started finding bugs and security issues. That’s fantastic! Independent review is exactly how complex software gets better.
There has already been an initial security review, and there’s a remediation plan sitting in context that I’m considering implementing.
So the question is, since the experiment is complete, do I even do anything further here?
Maybe there is an interesting experiment in measuring how efficiently human-directed AI can fix the problems external reviewers discover?
How much time does it take to reproduce an issue, identify the root cause, implement a fix, and validate the result? How much human intervention is actually required?
That’s potentially an even more interesting situation than the original 11-day experiment.
Because the real question was never whether AI could replace Linux. Even if it can replace some aspects of its functionality, it can’t replicate the community, standards, collaboration, alignment, or stability.
If you still believe that Ginux was designed to be a Linux replacement, you should read the actual paper.
zenodo.org/records/23682548
START retweeted
Agents are your new customers.
Introducing Intents by @tryscope_app
tryscope.app
Now you can see why they discover your product, and why they leave.
See what agents came to do, where they get stuck, and what's missing from your docs, content, and features.
We draft the fix. You hit publish. Then track the impact.


































