Kemble Song

Ordinary Human

Building a startup

The year is 2025. The AI gold rush is in full swing.

I decided to leave my cushy tech job to build a venture-scalable startup. After three pivots and countless rejections, here are my learnings and observations so far.

Don't do it (unless you have to)

Building a startup is a low expected value bet. Only one in ten actually succeed, and even that one will take five to ten years to materialize into any kind of value. VCs can justify it because they have hundreds of millions of dollars to deploy across hundreds of startups, where even a single unicorn out of those hundreds would more than cover the return on investment. Founders, however, don't have that kind of diversification. We go all-in on this one thing, committing our entire earning potential and a significant portion of our lifespan. The opportunity cost is brutal.

So why even do it? Entrepreneurship is an addiction. You can't help but scratch the itch. You've fully deluded yourself into thinking the only way you can leave a sizeable impact on this earth is to start a company—something completely new, that will hopefully create more jobs for the world and generate more wealth for humanity.

Some people can't help it. Many feel they will regret it if they never even try. I felt the same way.

Think bigger

We spoke with more than twenty VC firms, and nearly all of them gave some combination of the following feedback:

  • We like your team.
  • We like your vision.
  • Your raise is too small.
  • We're not convinced by your initial wedge.
  • Incumbents will crush you.
  • Fifty other startups already do the same thing as you (or close enough, or will easily incorporate it into their offering).
  • You should have opened with your pie-in-the-sky end goal—that excites us.

That last point really resonated with me. At the pre-seed stage, where you have no revenue and near zero traction, the only thing you can sell to investors is your pedigree and your lofty ambitions. You have to be braggadocious. You must have a non-mainstream opinion or a lofty vision of what the world will look like several years down the line. And you have to exude the confidence that you're the one to get us there.

We certainly had a lofty vision, but we chose to bury it as an afterthought to the product we had built to date. We also didn't seem very confident we could pull it off, given how little money we were asking for.

When you see how much hot startups are raising in early rounds now, and how quickly they're hitting astronomical numbers, most institutional VCs in this AI era are looking for founders who can think bigger and execute quicker. There is no room for the ordinary.

But also convince people this is something they want

The above section matters less when you can actually demonstrate traction through recurring revenue. I suppose this is the "zero to one" part that people say is the hardest, and I truly agree. Here are the steps to do this, distilled:

  • Find out what the "hair on fire" problems are for a particular function.
  • Find out how they are currently solving this problem.
  • Design and build something that solves it better.
  • Determine if "better" is actually worth paying for—both the cost of the software or service, and the cost of change.
  • If it is, congratulations: you've made something people actually want.

We had to pivot a few times to even catch a glimpse of real traction. We still haven't closed a deal yet, and to this day I'm unsure whether it's the "convincing" part or the product part that is lacking. They say any good salesperson can sell you a pen. Or perhaps this problem just isn't "hair on fire" inducing after all.

Don't sell to enterprise (first)

We were very excited when a large enterprise customer verbally agreed to sign up for a pilot. This would have been a six-figure, recurring deal—what B2B SaaS dreams are made of.

We got three weeks into their procurement process before we were ultimately blocked by their security team. It stung, considering how much work we had put in to get ready for the roll-out. From infrastructure hardening, to single sign-on integrations, to defining security policies and contractual agreements—every day brought a new question or requirement. We powered through this, even without a lawyer, relying on our own resourcefulness.

It felt incredible that our team of two was able to liaise with a 2,000+ employee company and their comprehensive legal and security team. But when it became clear to them that we were essentially "winging it" (they were shocked to discover our company had a grand total of two employees), they quickly shut us down.

It certainly hurt at the time. We spent so much effort getting the deal to that point, and on paper we had all the policies and controls ready to take it on. Even our infrastructure was scalable enough to handle their data load. In reality, it was naive to think we could land this deal. Enterprise is a beast, and above all else, what they're looking for is trust.

There's a reason why Microsoft, Google, Workday, and others dominate this area—they are household names insured up to the wazoo. Trust is a non-issue there. In reality, even if we spent the $20,000+ and four-plus months on getting a security compliance certification, and then the $10,000 on cybersecurity insurance, we'd still be short about four on-call engineers to support enterprise requirements.

We're just not at that level yet. Enterprise is the growth avenue, not the proving ground.

The bottom-up approach

Traditional wisdom for B2B SaaS startups is to rely on founder-led sales for initial growth. Get chummy with the buyers in the organization; wine and dine them until they're convinced they should spend money on your software.

As you scale, this translates to hiring account executives to do the same thing, and bada-bing bada-boom, you've got a growth engine. This is called the top-down approach, where a single person (or a small number of people) unlocks the budget to spend on software licenses and rolls it out to their entire company.

In the AI era, however, we're seeing something different. These insane startup valuations and growth trajectories are a result of viral growth. Viral growth happens when an individual loves using your product so much that they convince their colleagues to use it as well. Get enough people in a company to use it and, voila, you've inadvertently created an account executive who works at the company you're trying to sign.

Budget holders want to consolidate software licenses in their company, so when they see a sea of people using one particular tool, they'll opt to roll out the enterprise license. This is the bottom-up approach, which has led me to a few key insights:

  • This is product-led growth, which means your product needs to look and feel exceptional (people care about UX; companies do not).
  • This directly plays into the industry trend of software consolidation, where companies prefer everyone use the single best tool. The best tool is the one your employees are already using.
  • It's much easier to get one person to test your product than a whole company to pilot it.
  • With the AI boom, individuals are more open to trying new tools and reinventing their workflow.

Knowing this, my approach now is to lean more into optimizing for the individual. Once the network effect starts working, the top-down deal will sell itself.

Falling into the trap of fitting the tech into a problem

Getting a novel technology working and solving a real problem feels surreal. It's as if you've created something out of nothing, and it's going to start printing money. You get irrationally attached to the thing you've built. But then you find out this isn't something people actually want. So you try to find other problems where your technology could be used.

We fell for this trap when we figured out how to take transcriptions from meetings and turn them into coaching insights. This wasn't useful, so we repurposed our meeting tech for manager assistance, like writing follow-up messages. That wasn't useful either, so we tried using the tech for performance reviews. We got married to the technology, not the problem. Marry the problem, and the technology will emerge to solve it.

The end of SaaS as we know it

I now have a somewhat pessimistic view that general-purpose AI—with its multimodal chat and voice interface and tool use—is the ultimate user interface and will essentially destroy any software that is merely a usability enhancement over manipulating data.

Things that will survive (and thrive) are not really SaaS. They're repositories of data, infrastructure, or real-world actions that AI can tap into:

  • Marketplaces (Airbnb, Etsy)
  • Physical world interfaces (Uber, DoorDash)
  • Legal and financial infrastructure (Stripe, Plaid)
  • Social networks (Instagram, X)
  • Content, for now (YouTube, Netflix)

So the tricky question now is: What kind of software as a service can exist in this world, where ChatGPT and its ilk have all the knowledge in the world, can write code, and can even operate computers?

Perhaps the future of SaaS is just a license for AI agents to access your API...