Visiting Partner @ycombinator. Backing high-output YC founders via @phosphorcap. Previously founded and sold @ZeusLiving (S11) & @auctomatic (W07).
San Francisco
Joined July 2006
- Tweets9.5K
- Following4.1K
- Followers286K
- Likes12.4K
Pinned Tweet
Joining @ycombinator as a Visiting Partner for the Spring 2026 batch.
Feels a bit full circle.
Kulveer retweeted
One of the most unexpected things about YC is how long we continue helping the founders. Even though we're generally the first investors, we're the ones still advising them 10 years later. I think it's because we're the investing analogue of old friends.
One week left to apply to YC Winter '27!
YC founders get:
🔸 $500K in funding
🔸 $12M+ in tokens, credits & deals
🔸Partners who were founders first, and are in your corner for life
🔸 Access to the best founder community in the world
🔸 Demo Day in front of thousands of top investors
Apply by Nov 2: ycombinator.com/apply
Kulveer retweeted
We just launched TrueMile (@ycombinator F26), an AI-native trucking company.
Our AI plans loads, finds freight, negotiates with brokers and dispatches drivers. When it picked our loads, we earned 17% more per mile than the market.
Kulveer retweeted
Yep Most people Even in the industry do not realize how much demand there will be…
People should build more launch vehicles. There is an enormous sucking sound from the market that justifies many many efforts on this front.
Orbital compute will be >one trillion $ per year capex spend by 2035. There will be more than one launch company serving this. It’s not too late to get in the game, now is the time to build launch! 💪
Kulveer retweeted
never been more bullish on what we’re building 🥹
we recently crossed 10k+ customer accounts managed with @quivlyai
when we started, we had a strong belief that software will be bought and sold differently in the post ai world.
after analyzing thousands of accounts, it’s clear conventional playbooks are dead:
1. silence, not complaints, kills accounts: accounts with 0 support tickets have a 6.1% churn rate. accounts logging 6–15 tickets? 0% churn. people don't submit tickets for software they’re not using
2. more calls often means a rescue mission is in progress: accounts set to cancel logged a call 29.2 days ago - more than 3x more recent than non-canceling accounts (96.4 days)
3. fast onboarding (without enablement) is overrated: accounts taking 90+ days to implement and enable users stay 32% longer (477 days) with the highest health scores than the ones activated in under 30 days, often light on enablement. time to value still matters - it just doesn’t stand the test of time without real enablement
4. escalations save accounts, polite praise usually masks churn: accounts with logged escalations saw a 0.0% cancellation rate. accounts with praise threads? 4.2% churn. escalations mean leadership cares enough to fight for the outcome
5. single champions are a trap: moving from 1 contact to just 2-3 contacts drops churn from 5.9% down to 1.1%.
no human team can spot these quiet, counterintuitive signals across thousands of accounts before it’s too late.
good news is they don’t have to anymore. quivly does it for them :)
Kulveer retweeted
qwen's reasoning said it wouldn't call any tools. then it tried to upload a file to a beijing cloud bucket (sandbox caught it). i think this is why you can't secure agents by reading their reasoning. watch what they actually do with their tools
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Kulveer retweeted
ALT A graphic showing all English Ballon d'Or winners: Sir Stanley Matthews Sir Bobby Charlton Kevin Keegan Michael Owen
Kulveer retweeted
We Applied 2 times.
Best Decision to join and move to SF.
@ycombinator and our Partner @kul are pushing us really hard and we couldn’t be more thankful!
One week left to apply to YC Winter '27!
YC founders get:
🔸 $500K in funding
🔸 $12M+ in tokens, credits & deals
🔸Partners who were founders first, and are in your corner for life
🔸 Access to the best founder community in the world
🔸 Demo Day in front of thousands of top investors
Apply by Nov 2: ycombinator.com/apply
Kulveer retweeted
I'd stop running coding agents in ~/Downloads. Forever Security blocked a customer's agent from installing a malicious npm package. The order came from ~/Downloads/AGENTS.md. Their telemetry says about half of customers do this. That scares me.
Kulveer retweeted
There is something really concerning (and disappointing) going on in the US right now.
A dangerous and preventable situation.
Let me explain.
nextstrain.org/avian-flu/h5n…
1/
Kulveer retweeted
You misspelled “Russia”.
Kulveer retweeted
I'd check every ML repo for code relying on SageMaker's default bucket. Older versions used a predictable S3 bucket name without checking the owner, so whoever registered it first could swap your training code and take the role's credentials. Only if your bucket didn't exist yet.
why you need @Armorer_Labs
SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios.
The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios.
Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said.
Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior.
Read my latest for Axios here: axios.com/2026/10/22/ai-agen…
Kulveer retweeted
Using Claude Code or Codex agents to automate a business workflow is like asking Newton or Einstein to solve it from first principles every time.
The AgentRun harness we built is more like an experienced analyst: it learns on the job, takes notes, and modifies its workflow after every run. That means runs get faster and cheaper over time.
Day 4 of open sourcing parts of our harness. Today we made significant improvements to our workflow DSL. We are using it in our self-improving loop - turning agentic loops into more deterministic recipes.
With these improvements (a lot of @typesafe's Jev in routing, classification, and workflow composition), we have been able to reduce the cost of agents significantly without sacrificing performance.
I meant to open source the learning loop as well, but removing the business logic from @grepdotai still needs work.
Enjoy the demo tho : )
Kulveer retweeted
Introducing Orgo: computers for agents.
Give every agent its own desktop in the cloud, ready in seconds and always on.
Already used by 1,000+ companies, and launching today with @ycombinator.
Agents are creating a revolution in how businesses operate.
They also create new surface area for hackers to infiltrate and steal valuable data.
For this new world, @Armorer_Labs has built a continuous security layer that fixes issues without interrupting the business.
Super excited to have funded @begimher and @cristian_le0 in YC's F26 batch!
They built the first AI security agent at AWS, and are bringing their expertize to the world of regulated companies.
Check it out:
armorerlabs.com/
Today @cristian_le0 and I are launching @Armorer_Labs
It finds real security issues in the AI agents that regulated companies build, and fixes them without interrupting the business.
If you lead security or engineering, book a demo at armorerlabs.com
Kulveer retweeted
1) Today we're announcing Atomwell, our state-of-the-art foundation model for predicting experimental outcomes, alongside scaling laws for molecular discovery.
We trained models from 800M to 9B parameters on ~80M experimental measurements, and found clean scaling laws in both model size and data.
Kulveer retweeted
Roadster xitter.gq/i/broadcasts/1yKAPwNgg…
Congrats on your YC launch RATLOG @_charlieweise!
An extremely fast moving, resourceful team I was very excited to fund for the F26 batch with @ycombinator.
In 4 months they went from first build to a hot-fired engine, spending under $30K.
$400M in launch reservations already signed (so much demand that they skipped the usual LOI step that hardware companies go through).
We are building reusable heavy-lift rockets to take on SpaceX, powered by an engine designed with 90% fewer parts. First engine hot-fired 4 months in, from a basement, for under $30K. $400M in launch reservations signed. Satellite operators: let’s talk @ycombinator
Kulveer retweeted
We are releasing Ultramarine 1 Flash, a frontier-class model built for knowledge work. It is open-source and achieves similar performance to Fable 5, while being 60x cheaper.
It was post-trained by a team of 5 for ~$300k all-in. Available on API now, with weights releasing in 2 weeks.