Name: Sonya Huang and Sarah Guo on what's next in AI | Mercury Spheres
Uploaded: 2025-07-13T00:03:58.000Z
Duration: 39 min 21 s
Description: Thousands of YouTube videos with English-Chinese subtitles! Now you can learn to understand native speakers, expand your vocabulary, and improve your pronunciation...

All right, I'm super excited about this panel.

We have Sonia Huang from Sequoia Capital.

Sonia recently invested in Mercury, so obviously she's most excited about us.

But she's also a prolific investor in AI.

She did OpenAI, LangChain, Fireworks, and lots of others.

And we're going to spend this panel talking about AI mostly.

She's an early stage investor and has backed Runway, Cresta, Dust, and lots of other fun companies.

Also, she's a co-host of a podcast called No Prize.

So just to kick it off, can each of you share an AI investment you made recently, and what got you excited about it?

An AI investment I made recently is a company called LangChain.

They're a developer platform for building LLM applications.

And so if you think about what makes large language models very different, they're stochastic in nature.

Instead of code in, code out, oftentimes it's prompts that are driving these.

And as the agents that people are building grow increasingly complex, you actually need really robust orchestration mechanisms for doing things like if the agents can be long running, and working in the background, and coordinating with networks of agents.

And so LangChain was founded by a guy, Harrison Chase, who's been, I think, earlier to the agent trend than anybody.

We've been following him since way, way pre-Chat GPT and just following his view on how agents would shape up.

And he's very much building the scaffolding and developer infrastructure to support that world.

So yeah, we invested a couple years ago now, actually.

Sarah, you're also an investor in LangChain?

You had a fun story about how you got to be an investor there, right?

But one of the last times I wrote code somewhat involuntarily, I think it was a reasonable expectation for any entrepreneur to be like, use my product if you want to talk to me, investor.

And actually, speaking of financial visibility, I built a small LangChain application, which was essentially like a talk-to-your-finances app, where I wanted to get payroll data and cloud consumption data and just make it easier for early-stage founders in our portfolio to understand what was changing costs month to month.

But better mercury than me doing this work.

I think it's a fun story, because as entrepreneurs, we're often trying to get investors.

But it goes to show you that investors, when they really want you, they'll go write an app for you.

He made it a, I don't know if Harrison remembers this story, but he made it a barrier to taking a meeting.

To go back to the question, any interesting AI investments you've made recently?

Well, one company that I'm excited about, that we haven't talked much about publicly, is this company called Sola.

And so I think there's a lot for, I'm sure there are technical and non-technical entrepreneurs in this audience.

There's a lot for developers to do with language models.

There's also many processes in the world that feel very manual and unscalable, that non-technical people have the domain understanding for how to do.

And that could be the middle office in a financial services firm.

It could be operations in a e-commerce company.

But what Sola does is give you kind of like a, it looks like a drag and drop flow chart of processes, where you plug in language models to do actions in software that were very manual in your business operations.

And so I get really excited when other people get really excited about AI.

And so I was doing a customer reference for this company, Sola, recently.

And it was a guy who runs a billion dollar revenue logistics company, like a trucking brokerage company.

And he's like, we are transforming our business this way.

And there's no engineers in this company.

They have like one director of IT person.

And so I'm really excited when you are making this stuff more accessible to people.

I guess who in the audience who's an entrepreneur thinks of their company as an AI-first company?

Maybe put up your hands just so we get a sense of a smaller ratio than I would have thought.

I guess as you think about building an AI startup, it feels like the models are doing more and more.

Where do you think a founder can operate safely and interesting for you both as VCs to invest?

I'll start, because we've actually updated our thinking here.

So I would say, if you had asked me two years ago, I would have said, anything in the application layer, I believe that you're safe.

And our belief back then was that the foundation models were going to be so focused on research and building models in a lab that they just didn't have the time and the energy and the expertise to really focus on, how do they make their models great for any specific end customer use case?

And so I would say, overall, two years ago, my bias to that would have been, as long as you're building in the application layer, you're safe.

I think that mental model has changed a lot, in part because of just seeing the success that ChatGPT, for example, has had.

I think a lot of the application layer is actually going to be owned by the foundation models going up the stack.

And as a result, I think that if you're building anything horizontal consumer enterprise, I think that's right in the strike zone for the foundation models and where they want to go.

And encoding is also right in the strike zone for something they view as a critical path, a critical node to their missions.

And so those categories, I'd say, are relatively tougher to win in.

But then when it goes to, Sarah mentioned a few of these use cases.

We see this all the time of your health care, your hospital.

You really, really, really want just great medical transcription and billing.

That's just a different use case than the foundation models are building generally.

So I think if you're still building very vertical specific domain driven companies, that's probably where we're most interested in investing and where we think those companies are safest, so to speak, from the strike zone and where the foundation models are playing.

Sarah, did you have something to add to that?

Yeah, I was going to say, I feel like the last mile into every use case in the world and all these different professions, all these different industries is actually very long.

There are many things that are like earned expertise and trust that are not just intelligence.

I'm like, all right, a lot of entrepreneurs with high IQ.

And yet, if you go into your other's domains, there's no guarantee you're that successful day one.

I will never know as much about financial services innards as Ahmad does.

And that feels like very important to the success of Mercury at this point.

I'm sure it's kind of painful to climb that hill of knowledge.

But therein is kind of the advantage, right?

And so I think, especially in spaces that are operationally complex, that have regulatory concerns or where, honestly, the audiences, they are rightfully so a little bit skeptical and pretty concerned about the quality of the knowledge that's available.  let's say Sequoia and Conviction are firms, we're both invested in this company called Harvey that serves lawyers and a company called Open Evidence, which is medical search for doctors, right?

I use Chad GPT to educate myself about what, my kid's snot is purple today.

As like a first line defense, because it's hard to get a hold of a medical professional.

But if somebody is sort of trying to understand the most up-to-date research for a complex oncology case, I don't really want that doctor asking Chad GPT.

I want them looking up intelligently research and work by other doctors and medical professionals.

And so I think that there are domains where the specific data you're surfacing and the expertise and the trust just feels very different from some of the consumer applications we'll all use.

Subtitles ListPlay Video

Sonya Huang and Sarah Guo on what's next in AI | Mercury Spheres

specific

sort

bunch

mental