Thin and ephemeral vs. big and weak vs. small and strong.
Gil Dibner

It was only 11 months ago that ChatGPT was unleashed on the world, ushering in the GenAI gold rush of startups. This is an exciting and frightening time for B2B founders. While the possibilities enabled by these new technologies appear to be unbounded, the landscape is shifting so rapidly that it’s difficult for even the most experienced founders to pin down a direction that might hold water for long enough to build a business. The landscape (technology, infrastructure, and competition) is just moving too fast right now. As others (Sam Lessin and my partner David among them) have commented, this cornucopia of new technological possibilities does not translate into a particularly fertile landscape for early-stage founders.

Looking back on the past year, I can categorize many of the application startup ideas we are seeing into three broad groups: (1) thin and ephemeral; (2) big and weak; and (3) small and strong. (Infrastructure companies do not easily fall into these buckets, so let’s leave them out of this analysis.)

Thin and ephemeral. Most of the GenAI/LLM-based startup ideas we encounter fall into this category. They are characterized by an attempt to solve a very narrow business (usually horizontal) problem with nothing more than a prompt to an LLM, usually a third-party API. Many people refer to these as “thin wrappers” on LLMs. They are “thin” because they don’t really contain any technology (or, even, software) of their own. Just a GUI and a few prompt calls. I think of them as ephemeral because even when they manage to create some value, they are simply too easily ripped and replaced by the next best thing. That next thing could be a better/cheaper competitor, a feature on a broader application platform, or even a homegrown version of the same capability developed in-house using any of the readily available tooling. Almost by definition, the GenAI/LLM paradigm entails lower onboarding friction and lower stickiness. And in the vast majority of cases, the founders of these “thin” companies can’t articulate the defensibility of their company, because it’s built entirely on data they don’t own and the (hidden) computational frameworks of an infrastructure they don’t control or understand. These may be a quick way to make a buck from a customer or ten, but they don’t seem like sustainable businesses in most cases.

Big and weak. This second category is far more interesting intellectually. It’s seductive. The ideas are big, but they are so big that they are all convergent. When I reflect on the companies in this category in our deal flow (and there are many!) they are all slightly different versions of the same company. The headline claim ranges from “organize all your knowledge” to “build/integrate any application” but they all sort of converge towards the same idea. As the pitches go on, the ideas always keep getting bigger until — by the end of the meeting — it’s apparent that the founder is planning to ingest essentially all enterprise information sources (emails, logs, code, configuration files, salesforce records, SaaS APIs, you name it) and intends to enable all the users (business users, developers, and everything in between) to do just about everything from asking a question (a la ChatGPT) to creating a full-blown application (Github Copilot but on a massive dose of steroids and redbull). You may read this paragraph and think I’m referring to one particularly ambitious company, but I assure you I am conflating at least two dozen companies that promised all of these capabilities in the course of a single pitch. As a VC, I don’t know what to make of this. I assume some version(s) of this are possible, but the idea is now so commonplace that I suspect this is going to become an arms race as multiple groups compete to build this “everything” app. In that context, the $100M rounds we’ve been seeing in the space might make sense, but a team would need an incredibly strong set of claims as to why they are positioned to win that race and, crucially, how they would maintain their lead if they did (defensibility). That’s a tall order, and I haven’t yet seen a compelling case. So the pitches I’ve seen here are big ideas, but — unfortunately — too weakly expressed to be backable.

Small and strong. The final category of companies that we are seeing in this area are rare and very interesting. These startups are “small” because there is usually a clearly defined ICP, almost always in a vertical where the founders have deep domain expertise. This focus means that their TAM is constrained — large but not infinite. The scope of the product is also constrained. These products do not promise to do everything, but they promise to do some very specific things very well. These startups are often “strong,” in my view because they tend to have a pretty robust layer of application functionality built on top of the data and AI layer that underlies them. They sometimes generate their own proprietary data. They have enough domain expertise, for example, to be able to create barriers to entry around LLM inputs and outputs in ways that less experienced teams would not be able to devise. Their strategic depth often extends into their go-to-market operations. They sometimes have relationships and specialist team members who know how to penetrate the specific industry in which they operate. In short, their verticality is their key source of product clarity and business defensibility. These companies utilize AI and often have LLM capabilities as well, but these are mere building blocks that help underpin a much deeper and more robust application.

Needless to say, we are excited about these small and strong companies. In a classic venture sense, they are, of course, not “small” at all. Their markets can be enormous. But they appear small relative to their high-flying AI-flinging peers because they are not attempting to be everything to everyone all the time. LLMs are allowing founders to indulge in the fantasy that they can build everything for everyone all at once. In some way, they can. But these “big” product visions are likely to end up being weak products and even weaker businesses. Lately, our money has been on companies with a tight vision around how to solve the real-world business problems of a customer set they know intimately.

If you are building that, we want to hear from you.

FROM THE BLOG

The End of Entrepreneurship by Autopilot
The unicorn factory has stopped. What are the implications for founders?

No Words
The heartbreaking situation in Israel.

2023: The Crucible Year
What is in store for the next 18 months?

Kafka Survivors of the World, Unite!
Why we backed Memphis.dev.

WORTH READING

ENTERPRISE/TECH NEWS

Regulating AI. Last week President Biden issued a comprehensive executive order aimed at shaping the future of AI in the US. It outlined several key points and actions to ensure the responsible development and use of AI. In response, Bill Gurley shared his concerns that the AI incumbents, like OpenAI, are leveraging regulation for their own benefit — he believes that an executive order wasn’t necessary and the large AI companies shouldn’t be involved in the regulation process. “There’s plenty of experts, academics, whatever. They aren’t conflated with their economic interest. Go build your regulatory plan with those people. Don’t do it with the companies that are begging for regulation.” According to Bill, the biggest risk of the incumbents being involved in regulation is “if they succeed in making the open source AI models illegal. That would be horrific and profoundly impact innovation in the AI space and force everyone to buy from the big companies.”

WeBankrupt. Once the US’ most-valuable startup at $47B, WeWork is now planning to file for bankruptcy as soon as this week. WeWork’s move towards bankruptcy comes after significant turmoil, including its failed attempt to go public, a flawed business model and a reduced demand for office space. “The vast majority of WeWork’s leases were signed in 2018 and 2019, when rents peaked before the pandemic. Now, amid a work-from-home boom, tenant demand has plummeted in major U.S. cities such as New York and San Francisco, where WeWork is concentrated. As of June, WeWork was paying over $2.7 billion a year in rent and interest — more than 80% of its entire revenue, according to company filings. That was nowhere near enough wiggle room to cover the company’s other expenses and turn a profit, and a worse profile than competitors. Its total losses since founding topped $16 billion as of June, as it churned through all the money it raised from top investors and lenders over the past decade. Even after four years of cuts and reorganizations under new management that followed Neumann, the company was still burning through $300 million of cash a quarter.”

Guilty. Sam Bankman-Fried, founder of FTX, was convicted last week on seven charges of fraud and conspiracy, facing a maximum of 110 years in prison. SBF was once seen as the white knight of crypto, bailing out crypto companies like BlockFi and Voyager Digital. The trial highlighted the spectacular fall of both SBF and of crypto.

HOW TO STARTUP

Learnings from ServiceTitan. Alexandre Dewez, a VC at Eurazeo, shared nine key insights from studying ServiceTitan. The entire post is worth reading, but learning number 8 is especially worth noting. “Be the premium software in your market. ServiceTitan is the most comprehensive but also the most expensive software solution in its market. It’s 3x more expensive per technician compared to solutions like Jobber or HouseCall Pro which are going after SMBs. Moreover, ServiceTitan has launched Pro version of its modules to enhance this premium positioning. Being seen as premium helps you to become the market standard and increases the willingness of your customer base to refer your product.”

The SaaS trends report. Vendr has released their SaaS trends report for 2023 Q3. Some of the most interesting data shared was that AVCs have rebounded, “climbing 43% QoQ, but remains lower than three-year average” and still “17% lower than 2022 Q3”. Another interesting insight was the top five purchased categories of Q3 — as this shows where companies are still willing to spend in this market:

1. Business intelligence

2. Data integration

3. Data science and analytics

4. Cloud security

5. Cybersecurity

HOW TO VENTURE

Seed graduation rates. Precursor Ventures’ Charles Hudson shared his musings on the seed to Series A graduation rates plummeting. “Historically, we’ve seen a strong pipeline of companies moving from seed to Series A. Recent numbers, however, indicate a significant decline in this graduation rate. Measured graduation rates will continue to fall for several quarters as companies go out for and fail to raise Series A rounds. Graduation rates from seed to Series A could drop to 25%, or one-third or one-half of what they were at the peak.” What are the implications of this reset? Not only will VCs need to raise the bar for the companies they invest in even higher, but they will also need to “consider fundraising risk for companies that need a Series A to achieve their mission. Not every seed company needs a Series A round to succeed; more will have to succeed without additional capital.” Will the Series A market come back? Mostly likely, eventually, but it’s impossible to know when with certainty and “unfortunately for seed-stage companies, the Series A market can remain on strike longer than most seed-stage companies can remain solvent.”

PORTFOLIO NEWS

Viably, which funds Amazon sellers’ businesses, has secured $50M debt financing.

Jurnee’s CEO, Tania Kefs, spoke on the GTM Strategist Podcast about sales-led growth and how to do cold outreach when you are just starting.

CruxOCM’s CEO, Vicki Knott, shared some details about the renewable energy waste crisis.

Aquant’s CEO, Shahar Chen, wrote about a red flag every consumers should watch for before sharing their personal data.