Things I learned in my first year running a startup

A year ago, I didn't have a company. I had a hunch.

The hunch was simple: people were starting to ask ChatGPT things they used to Google. Not occasionally. Constantly. And it wasn't just "what's a good recipe for dinner" stuff, it was "what's the best tool for X," "which vendor should I use for Y," the exact kind of question that used to send someone down a ten-blue-links rabbit hole and land them on a landing page with a demo button. Except now the answer just... showed up. One recommendation. No links to click. No comparison shopping. Just an AI telling someone which company to trust.

I remember thinking: if that's where the recommendation happens, that's where the budget is going to follow. And nobody serious was building for it yet. Every "AI tool" I could find was a feature bolted onto something else: a screenshot generator with a fancier name, not a real product built around the question that actually mattered: are you the company getting recommended, or are you the company nobody mentions.

That was the whole company, really. One bet, made on a random week, before we had a name for it, a website, or a single line of code between the four of us. No deck, no investor pitch, no validation framework. Just four people who couldn't stop talking about the same problem, and the slightly stupid confidence required to act on a feeling before you've proven anything.

Three months, four of us, way too many calls

We didn't agree on much in the beginning except one thing: talk to as many people as possible before writing a single line of code we'd regret. Between December 2024 and February 2025, the four of us split up and ran our own interviews: different networks, different industries, different angles on the same question. Nobody compared notes until afterward, which turned out to be the smart part. We weren't trying to confirm what we already believed.

What came back was more useful than any of us expected going in. A pile of feedback, half-formed product ideas, objections we hadn't considered, and enough overlap across totally unrelated conversations that it stopped feeling like a coincidence. People kept saying some version of "wait, so you can actually tell me if I show up when someone asks ChatGPT about this?" That reaction, over and over, across four different sets of conversations, was the first real signal. Not a survey. Not a TAM slide. Just a lot of strangers leaning forward.

That's lesson one, and it's not original, but I didn't believe it until we lived it: conviction is cheap, four independent founders hearing the same reaction from people who owe them nothing is not.

March 2025: $500, and the first real proof

The first sale closed in March 2025. Five hundred dollars. Not a number that changes anyone's life, but it changed how the four of us felt about everything we'd been building. Before that, we had a product we believed in and a stack of encouraging interviews. After it, we had someone who'd actually handed over money for it, which is a completely different kind of validation than someone being polite on a call.

That's the moment it stopped feeling like a project we were testing on people and started feeling like something someone would actually buy. Small as it was, that $500 did more for our confidence than the three months of interviews before it.

Bootstrapped means everyone does everything

We didn't raise anything to get this off the ground. No seed round, no safety net, no investor update to hide behind. Bootstrapped, full stop. Which meant the four of us were writing code in the morning, doing outbound in the afternoon, and somehow also being "support" by evening when an early user hit a bug at 11pm.

I'm not going to pretend that's some kind of badge of honor. It's mostly just stressful. But it forces a specific kind of clarity, because every hour you spend has to justify itself against the hours you don't have. You stop building things because they're interesting and start building things because someone specifically asked for them, twice, and meant it.

Go-to-market had its own start-stop rhythm, which I've since decided is just normal and not a warning sign. Webinars worked, for a while (they brought in a real handful of early sales), and then, right as we leaned in harder, they stopped converting and we couldn't fill a room no matter what we changed. So we moved on and tested everything else at once: cold email, cold calling in both India and the US, LinkedIn outreach, a few early partnerships. Cold email turned out to be the most interesting one, mostly because we hadn't expected deliverability itself to be the real battle: getting an email to actually land in an inbox turned out to be a bigger problem than getting someone to reply to it once it did.

The problem inside the problem

Here's the part nobody warns you about: building a product to solve a market gap means you immediately run into the same gap yourself, except now it's your problem and not a hypothesis.

We needed to know, constantly, whether we were showing up when someone asked an LLM about this category. So I went looking for something that could function as a basic ChatGPT Rank Checker & Tracking Tool for B2B, anything that would just tell me, reliably, "yes, you got mentioned" or "no, you didn't, and here's who did instead." I expected this to exist already, fully formed, probably overpriced. It didn't. What existed was a pile of half-features bolted onto SEO tools that were never built for this.

So we built our own. That internal hack (three people, way too much coffee, checking the same ten prompts every morning like it was a sports score) became the seed of what eventually turned into real, sellable LLM SEO Tracking Software for B2B. Not a side project anymore. The actual product.

Then the second AI showed up

ChatGPT was the obvious first target, but it wasn't the only one. Within a couple of months it was clear Perplexity buyers behaved completely differently: more research-heavy, more citation-driven, more likely to actually click through to a source. Ignoring it would've meant optimizing for half the picture. So the checking tool grew a second engine, and what started as a single-purpose script turned into something closer to a real Perplexity SEO & Rank Tracker Software for B2B, because the two platforms simply don't reward the same things and pretending otherwise would've been lazy.

Knowing isn't the same as understanding

Tracking whether we showed up answered one question and immediately raised a more annoying one: why didn't we show up on the prompts where a competitor did? That's a completely different problem than "are we mentioned, yes or no." It meant digging into what the model was actually pulling from: which pages, which phrasing, which third-party mentions were doing the heavy lifting for someone else.

That's the part that eventually became a proper LLM SEO Analysis Tool for B2B inside the product, because a tracker that just shows you a red X without telling you why is mostly a tool for feeling bad, not for fixing anything.

Fixing it became its own job

Analysis told us what was broken. It didn't fix it. For a while, "fixing it" meant me manually rewriting a page, waiting two weeks, and praying. That's not a process, that's superstition with extra steps. So the same scrappy instinct that built the tracker kicked in again, and we ended up with what's now an actual LLM SEO Optimization Software for B2B layer, the part of the product that tells you specifically what to change, not just that something's wrong.

Why "good enough" tools weren't good enough

At some point I went looking at what else was out there, mostly out of paranoia that someone had already built this better than us. Most of it was the same story: a single prompt, a screenshot, a "visibility score" with no consistency behind it. Nothing built to be trusted week over week. That gap is exactly why we kept pushing toward being the best LLM SEO Checking Tool for B2B we could build, instead of shipping something that looked impressive in a demo and fell apart the second you ran it twice and got two different answers.

The grind nobody puts in the highlight reel

By the end of the year we'd grown from the four of us to a team of eleven. That sentence undersells what actually happened in between: the missed weekends, the customer calls that ran into dinner, the days where the only win was "nothing broke." Pure grind, no shortcuts. I don't think there's a clever trick that skips this part. Anyone who tells you there is hasn't actually done it.

Hiring our first person outside the founding four was harder than hiring our seventh, mostly because there was no playbook yet: no onboarding doc, no clear sense of what "good" looked like, just the four of us trying to explain a half-finished vision out loud and hoping it held together. By the time we were hiring person seven or eight, we'd actually figured out what we needed, which made every hire after that faster and a lot less stressful. Nobody warns you that the hardest hiring happens when you have the least idea what you're hiring for.

We also lost a deal early on that taught me more than most of the ones we won. A prospect asked for something we didn't have yet, we said yes anyway because saying no felt like losing, and we spent three weeks building a feature for one account that nobody else asked for. They still didn't sign. That was an expensive way to learn that "yes" isn't a strategy, it's just a way of avoiding a harder conversation about what you're actually building and for whom.

What I'd tell someone starting their own first year: the idea matters less than how fast you're willing to find out you're wrong about it. We were wrong about plenty: what to build first, who to hire first, how long things would take, which "yes" to say and which one to turn down. The only thing we got right early was paying attention to a shift that was happening whether we showed up for it or not.

A year in, I don't have it figured out. I have a team, a product that actually works, and a much longer list of things I still don't know. That feels like progress.