From 100K to 200K users in 18 days

We set ourselves a goal back in April of 100,000 users by the end of September, and 200,000 by the end of the year, and honestly both felt pretty ambitious when we wrote them down. We crossed 100,000 on September 10, twenty days early, and we were still working out how to celebrate when the next 100,000 arrived 18 days later.

As I write this, we're past 240,000 already, and the counter hasn't really slowed down.

We've spent the last few days going back through what those three weeks actually looked like, from the people who showed up to the machines that kept up with them, and we wanted to share it, partly because the numbers are genuinely fun and partly because they say a lot about where we are as a company right now.

What people handed the agent

For the last few years, asking an AI for help with something on the web mostly got you instructions, and the clicking was still yours to do. What September showed us is how many people have stopped settling for that. They ask their AI to do the job itself, and when the job lives on a website that doesn't have its own connection to the AI, TinyFish is what it reaches for to get it done.

What surprised us most was how specific people got. An eBay seller asked it to open Seller Hub and list every order that came in that day, with payment and shipping status for each one. Someone setting up a company had it build their entire Notion workspace, from the sales pipeline down to the SOPs, and a shopper had it read through the reviews on Amazon's best-selling robot vacuums and pick the right one for a home with pets and hardwood floors.

None of these are glamorous, and they're exactly the kind of jobs people used to lose an afternoon to.

The requests that told us the most, though, came from people letting the agent into the tools they run their businesses from. One marketer had it set up and publish a new Meta leadgen campaign, budget and all, without touching any of their existing campaigns. Someone else had it pull the month's profit and loss report out of Xero because their bank was asking for it, and others pointed it at their Shopify admin, their Cloudflare dashboard and their Vercel projects. Almost every one of those requests came with its own guardrails written in, like “read only” or “don’t change anything else,” which tells you people are thinking carefully about what they're trusting it with.

When someone lets an agent into the place their business runs, it has to be safe as well as working, which is a big part of why we shipped Vault and Profiles before the surge.

It also came from everywhere at once. The traffic stopped having a quiet time of day, and even the slowest hour during the surge was busier than our busiest hours used to be.

What it looked like from the inside

In those three weeks, our Web Agent took on nearly 16 times as many runs as in the three weeks before, and every one of those runs needed a real browser.

If you add up the time every agent spent working in those browsers, it comes to about six years of browsing, done in just three weeks.

We kept outgrowing our own infrastructure along the way, raising a limit, hitting it again and raising it again, with new capacity coming online in the middle of the surge.

Which leaves the question that matters to anyone using TinyFish, which is whether any of it held up.

The numbers we're proudest of

Mostly, the ones that barely moved. Spinning up a browser took a median of 75 milliseconds the entire time, exactly what it took in a quiet week in August. And while the front doors to our product took more than twice the traffic of the three weeks before, errors on our end got four times rarer. When Pasha Dudka, our Head of Engineering, wrote up the three weeks for the team, that was the number he asked everyone to notice: “Four times better, at double the load.”

It wasn't perfectly smooth, and we'd be lying if we said otherwise. We had a handful of long days where parts of the stack needed more room than they had, and the team stayed on them until they had it. A lot of why it held comes down to work we'd done long before the surge.

The work that paid off in September

Almost everything that kept TinyFish standing in September was finished before September started. The serving stack for TinyFish’s web native model, the work on the agent harness, model routing and fallback, a rebuild of how tasks get picked up, instant browser provisioning, Vault and Profiles, and the fixes that made our MCP server install cleanly all landed in the months before. We'd been building for the moment people started handing real work to agents, and when it arrived, all of it held.

Our CTO, Mark Peng, described that kind of work to the team a week into the surge, and it's the best summary I've heard of why September went the way it did: “It’s about building the engine so that your results compound on one another, and the trajectory matters more than any single moment.”

The other thing that held was the team. Nobody froze the codebase, and the busiest day for merged code in all of September turned out to be the same day as our biggest traffic day ever.

Pasha called the fixes during the surge “the boring kind that keep a company alive,” which meant raising a ceiling here and widening a connection pool there, and plenty of them went out well after midnight. It was the same story across product, growth, revenue and marketing, everyone moving at that pace together.

A lot of that pace came from the PLG team, which Kat McMillan and Kris Szeto, our PLG Leads, pulled together in late April around the original 100,000 goal. Kat wrote about how they approached it, and her version is shorter than mine: “The team had goals. Our goal was the team.” Zach Lyon, who spent months taking apart our onboarding step by step, came away with the lesson I'd want any team to borrow from us: “You cannot optimize a funnel you don’t understand as a user.” That's the part of September I'll remember longest.

We're just getting started

We hit our year-end goal three months early and we're past 240,000 now, and the thing we're most excited about is that the systems and the team underneath that number feel more ready now than they did when September started.

Florence Le, our Head of Marketing, put it best when she told the team the plan had changed: “We were going to celebrate 200K on Oct 1. Then the numbers kept moving. We'll be at about 240K by launch day, so a 200K party would already be old news.” We have a lot more shipping this month, and our next goal post is 300,000.

We'd love for you to come along.

Kris is already looking past the numbers to the people behind them: “There’s so much we can and should be doing to make this a seamless and magical experience for first time users.”

To say thank you to everyone who got us here, we're adding 30% extra every time you add money to your TinyFish account through October 31, and you can do that right here. And whatever you've handed the agent so far, whether it's a Google Form or your whole Shopify admin, tell us about it and tag @Tiny_Fish.

We read everything.