Enterprise AI VC @FoundationCap | Early investor in @databricks @tubi & 6 other unicorns- @cohesity @eightfoldai @turingcom @amperity @alation @anyscalecompute
There are so many others to thank but we also want to give a shout out to the following people:
- Abhi helped build our inference product from the ground up and has the best customer empathy I have ever seen.
- Jared Rosner from Elad Gil’s fund literally forward deployed with us for months to help us figure out how to procure compute reliably.
- Jaya Gupta taught us how to think about what enterprises cared about most, helped us hire incredible people, and tapped into her amazing network for customers intros.
- Josephine Chen from Sequoia was one of our earliest partners. She helped us find customers when we had no product.
- Connor Ling helped us recruit and close AMAZING talent. Many of our top people came through Connor’s referrals.
⁃ Rachel Park from Benchmark was an amazing friend, recruiter, and helped us with day 0 business operations when we had nobody.
Early stage founders: fine-tuning models can wait.
At pre-seed, it’s not the time to debate Michelin vs McDonald’s.
You’re tight on capital. Eat your nuggets and enjoy.
As an early stage founder, your sole job is to identify a customer pain point and solve it. But you have limited runway, and you’re on a ramen budget. So you need to be thinking about the lowest cost per gram of protein: the most efficient path to solving that customer’s problem.
That’s not fine-tuning open models.
The benefit of building on open-source models is that you can customize them to your exact requirements instead of taking what comes out of the box from a frontier lab.
But fine-tuning takes money and time. You also have to host and run the model yourself. And most of the time, an off-the-shelf model gets you 90% of the way. That’s usually all you need to start.
The rare exceptions are cases like medical diagnosis, where a small accuracy gap can mean missing a real condition. So fine-tuning model accuracy by a couple percentage points matters.
But for most other early stage founders, you should not be fine tuning until much further along.
New: Building agents on @AWS Bedrock with data security across agents, models, and datastores.
Exciting to be doubling down on our partnership with Amazon.
https://t.co/tPLXCEya6N
One of the biggest mistakes founders make:
They hire their first employees just to fill gaps.
You need the front end done, so you hire a front-end engineer. You need to start selling, so you hire a salesperson. Hiring becomes a way to take work off your plate.
The problem with that - you’re hiring for immediate relief. You evaluate candidates on whether they can do the junior role today, without considering what they need to grow into. Your startup pays for that short-term thinking later.
When it’s just you and your co-founder, your first two hires help set the standard for the next two, and then the next two.
You need to hire with the expectation that your early employees will become your company's future leaders. Your next VP of engineering could be one of the junior engineers you hire today.
That means early hiring decisions carry immense weight. You need to look for signals that candidates can grow with the company and that they will be bar raisers who can eventually lead.
Before you make an offer, ask yourself: "A year from now, would I trust this person to build and lead a team of their own?"
Open Source has gone from zero to 30% mkt share in 5 months. OpenAI gaining share in enterprise. What does this mean for Anthropic’s IPO ? And why is Google still MIA ?
We benchmarked Genie, powered by Genie Ontology, against two frontier agents on enterprise data tasks. Here's what we found.
In collaboration with @turingcom, we developed a benchmark that tests more than table queries. Agents had to use business knowledge spread across notebooks, dashboards, and saved SQL queries in a live Databricks workspace, running each agent five times per task.
Genie passed 91% of runs, compared with 45% and 58% for the other agents, while using less than half the tokens and finishing 3–4× faster.
The key difference is business context. A valid SQL query can still produce the wrong answer if it uses the wrong definitions or calculation rules.
Genie Ontology retrieves and prioritizes business knowledge, helping agents find and apply the right context instead of relying on table schemas alone.
Explore the benchmark and methodology: https://t.co/q5xup45puH
Open Source has gone from zero to 30% mkt share in 5 months. OpenAI gaining share in enterprise. What does this mean for Anthropic’s IPO ? And why is Google still MIA ?
2/ Provider diversification: Enterprises are increasingly using models from more providers. At mid year most companies had one model provider. Today, 75% have two or more, and almost half use three or more. The most common "new providers" are OpenAI (GPT models) and OSS models.