From T2 to Pebble: The Rise, Challenges, and Lessons of Building a Twitter Alternative
Tl;dr We tried to build a Twitter alternative from scratch, starting a few weeks after Elon Musk took over Twitter and laid off half the company.
Initially we called it “T2”, and we later renamed it to “Pebble.” We raised $1.4M from angels in December 2022. We grew it to 20k registered users who have posted 350k posts. At the peak we had 9k MAU and 3k DAU, before we stopped growing. Our traction was insufficient to raise more venture capital, and just as we were going into our fundraise, our DAUs started shrinking. We decided to end our experiment and shut down Pebble on November 1, 2023.
This is what we learned along the way.
T2 in Screenshots
2 weeks to a first launch. We launched t2.social a mere 2 weeks after the first git commit. We chose to constrain users initially, and at the end of the first day, we had 6 users. This is what it looked like. Notably the “Notifications” tab didn’t actually work:
The Redesign. My cofounder Sarah and I had worked at Twitter before starting Pebble. We ere able to convince two very senior ex-Twitter to help us redesign T2 into something more visually appealing. We decided to build something that looked retro and whimsical. I think this redesign was a big success: There was a small minority of folks who hated the left drop shadow, but most people liked the design, and the retro look played well with our “Twitter, but back to the roots” message.
T2 had been the first name that came to mind, and our goal was to eventually find another name that wasn’t that derivative. It took us a long time to secure a new domain and trademark, but eventually we settled on renaming to “Pebble”, with the URLs referencing the fact that the people make the platform: E.g. my profile was at pebble.is/gabor
In Numbers
We used Posthog for metrics tracking, and I highly recommend it, even if it’s a tad bit expensive — our Posthog bill was about $500/month towards the end. Here’s our DAU chart over time, marked up with key events:
Most consumer VCs will tell you that you need at least 3k DAU in order to qualify for a Seed raise. We were above that minimum bar for 3 days of the product’s existence: the 2 days of Handlegate (see below), and when we announced our rebrand from T2 to Pebble.
Next, below our week-to-week retention chart. This is a very common metric for investors to ask for, as it shows if your product is sticky with users. Specifically, it shows what $% of users that joined in a given week came back to the product N weeks later. Each line in the chart is a week of users joining.
What’s the zig-zag line on the retention chart? That is Handlegate — more details are below, but this was basically an accident in which I sent out an email to users announcing that we’ll take back their handle. That caused users to return across every cohort we ever onboarded. The zig zags are exactly one week apart. The most recent cohorts were not affected by Handlegate, as those users joined after it occurred.
Finally, here’s a graph of cumulative registered accounts on the platform. (Keep in mind this is a graph that can only go up.) Pebble is ending with just shy of 20k registered users. On this graph, right around June 2023, you see a the start of a growth period as we are opening up to folks on the waitlist. Eventually we exhausted the waitlist of about 33k people, and growth flatlined. In September 2023, we announced the rebrand to “Pebble” and growth kicked in again, then flatlined as we failed to go viral.
Why did we fail?
The topline reason for the shutdown was that we couldn’t figure out how to grow. Why couldn’t we grow?
- There was a glut of alternatives, and we couldn’t clearly articulate a differentiator beyond trust and safety.
- Being kinder and safer is not enough of a differentiator, because kindness and safety by themselves aren’t interesting enough. There was not enough interesting content on the platform for the experience to become a daily habit.
- I mistakenly didn’t prioritize building an app, which would have helped with retention, or an API, which would have helped with pumping more content into the platform.
- Renaming from T2 to Pebble helped our perception with investors, but not sufficiently to warrant the tanking traction.
- We never had a growth strategy beyond the press-driven waitlist. We expected that viral growth might kick in when we open up more broadly, but it didn’t.
Things We Did Well
Despite the ultimate outcome, I do think we did a few things well. Here’s what I think they are:
Launch Now. We launched the first version of T2 in 2 weeks. My very first startup was a YCombinator startup, and I want to follow their Essential Advice. At the top of their list is “Launch Now” — just build something simple and get it out right away. When the T2 effort was first written up by TechCrunch’s Ingrid Lunden, we launched a waitlist which quickly grew to 10k+ people. I got to work on the site right away, and launched it two weeks after the first git commit, on Dec 9, 2022. But we had to constrain the problem somehow, so initially we just launched T2 with just 6 users, all of them people involved in the effort. The fact that we already had a publicly visible product, and we weren’t just vaporware, increased interest in T2.
Quick iterations with user feedback. The first versions of T2 were pretty bad. We decided early on that we were going to be super responsive to our early users and fixed bugs as they were being reported. This showed users that we care, and I think in the end, our website was really good. We were determined to ship a new version of the site once a day every, Monday through Friday. We @-tagged users who had reported bugs in the release notes. It turns out there’s still a real community of early adopters who want to be part of early social networks as they emerge. They want to contribute and help, and want to have a connection with the founders, and are more than willing to help.
Press. YCombinator tells you that press is not a good way to grow your product, because getting press is not repeatable. Initially, this didn’t turn out to be the case for T2, which had a strong A vs B narrative (David vs Goliath). The press loved writing this story, and there wasn’t a single week where we weren’t at least mentioned in a piece. Nearly all traffic bumps occurred on days where we had an article written about us. In a few cases, we had articles written about features we launched without having pitched the story, e.g. this piece in TechCrunch about DMs, and a bunch of coverage we got for our launch of the algorithmic feed. In January, we hired Heath Fradkoff at Ward 6 who was very helpful managing our press relations, and if you have a startup you’d be lucky to get to work with him.
Perhaps in the end, YCombinator’s advice to not rely on press as a growth engine turned out to be accurate. We didn’t have a great growth strategy in place for when the press ran out.
Trust and Safety. We decided early on that we would prioritize trust and safety on the platform, and had a latecomer advantage. Trust and Safety something that Twitter seemed to be moving away from. My cofounder Sarah came from a T&S background: She joined Facebook to help address the Myanmar and Cambridge Analytica crises, and at Twitter she had done a lot of policy work. Super early on, a bunch of people told me that Trust and Safety was not really solvable, and that trying to do so as a small startup was a futile undertaking. I think this is not accurate. Moderation has become more tractable: With off-the-shelf AI classifiers, you can easily spot posts that are in violation of your policies, and remove the content from the platform quickly. Misinformation is hard to identify, but it gets flagged by users very soon after it gets posted. You have to make T&S a priority early on.
Bluesky got a bad reputation for the unseemly content that got posted early on, and the vibes of Bluesky being an unsafe place have continued even once they introduced more proactive moderation.
Unsure if We Made the Right Call
Here are two decisions where I’m not sure if we made the right call:
The Fediverse — We had a few very vocal users really pushing for us to join the Fediverse. Meta’s Threads announced that they would do so, but they haven’t done it yet. Post.news announced they join, but I don’t see that in their product yet.
Joining ActivityPub would have been a big engineering effort, but federation could have brought in enough interesting content to make sticking around be worth it.
Our Handle Policy — We felt instinctively that users care a lot about having a short handle, or a handle that matches the one they have on other networks. Early on, we did two things:
- We found a list of the 424k legacy verified users on Twitter on Github, and we pre-reserved all those names. You could only claim one of those with a special invite.
- We blocked off usernames that consisted of swearwords, etc., though i was surprised by all the workarounds people can come up with to have questionable handles.
- If you chose a handle shorter than 8 characters, we asked you to select a “backup handle” that was a minimum of 8 characters. We set the expectation at signup that we would reclaim your handle if you went inactive on the platform for >30 days.
It turned out we were right — people care a lot about their handles. They ideally want their first name, or they want their handle to be consistent across platforms. Our intuition was right, but I’m not sure if we made it all more complicated than we should have.
There are few issues we brought upon ourselves.
Accident #1: Handlegate
In early September, I was setting up a first test for the reclaim handle mechanic. I drafted an email that I would send to 10 users that had long gone inactive on T2 (this was before we renamed). The emails looked like this:
But instead of sending this to just those 10 users inactive users, I fatfingered the “Recipients” field in the Sendgrid console, and sent the email to ALL the 12000 users we had on the platform at the time.
It turns out people really care about their handle, and this email caused so many users to come back to the platform so quickly and try to login that we had an outage. This also helped us uncover that we were doing too much compute-intensive stuff right after login as we were loading your timeline.
This is why in the charts we’re sharing below you’ll see a huge bump in usage and retention at the start of September: nearly 60% of registered users came back to the platform to keep their handle.
Accident #2: Ideageddon
Over the summer, I had experimented a bunch with suggested posts and replies, and in late September, we were ready to launch our “Ideas” tab, which would suggest new posts based on your old ones, and new replies to people you had previously replied to. This was all powered by OpenAI’s GPT API.
When we launched this, the reaction from the Pebble community was immediate — and negative. People felt that the AI was putting words into their mouth, and that the entire platform less authentic, and full of spammy replies.
We quickly un-launched this feature, and never really relaunched it. Style transfer from your previous posts to new posts remained hard to really get right, and the models also tended to hallucinate wildly when asked to generate new posts. Our cofounder Mike mentioned the fact that he owns a Jeep in his bio, and GPT would make up entire fantasy trips with his Jeep that he had never actually taken.
Could Have, Should Have
In retrospect, there are a few easily avoidable mistakes I made early on:
Should’ve Started Earlier: When I left Google in July 2022, one of the startup ideas I considered right away was building another Twitter. Even the pre-takeover, pre-X Twitter wasn’t a great place. But I hesitated since I didn’t think the takeover would go this badly.
If I had started building something the day I left Google on July 7, 2022, it could have been ready to onboard users by the time the post-Twitter opportunity arose. But instead of having a product, I had to put up a waitlist, and much of the interest in a real Twitter alternative was captured by Mastodon.
Should’ve Started with a Fully Algorithmic Timeline: We launched our algorithmic feed in early September, to a mostly positive reception. We had feared a backlash because some users had expressed a very strong preference for a reverse-chron feed that gave them the feeling of full visibility. While I can’t quantify the effect of introducing “For You” on retention due to the variety of other features we were launching at the time, my feed did feel more relevant, more alive, and more aligned with my interests than before. The logic that went into sorting “For You” was pretty simple. In retrospect, we should have never launched anything else. You’d think that in the post-TikTok world, just doing an algorithmic feed would be a no-brainer, but we didn’t get to it in time.
Should’ve been clearer about what kind of content we want to see: We never stated explicitly whom T2/Pebble was for, and what kind of content we wanted users to contribute. Users tried a variety of things, e.g. on one day a bunch of early users flooded us with memes. We had some general news, we had users on both sides of the US political spectrum, some sports, and we had a fairly active community talking about tech news.
We should have chosen one topic and a single set of users, and then put our initial focus there. My instinct is that perhaps we should have started with tech news: Naturally Twitter / X were a big topic, and we managed to recruit a bunch of tech journalists to the platform.
If we had articulated more clearly who T2/Pebble is for, and what content we want to see, we would have been able to boost that content algorithmically across all surfaces: Bump it up in the algorithmic feed, increase visibility in Suggested Follows, and reward posting about those topics with visibility and interaction.
Focusing on one type of content would have answered an important question we never answered: “How do you win on T2 / Pebble?” — if we had given a clearer answer to this question, I think we may have succeeded in gaining a real foothold in a niche, and then we could have grown from there.
The New T on the Block
The day Meta’s Threads launched was an interesting day. There was quite a bit of foreshadowing with the pre-launch leaks at Meta.
The day Threads launched, the frenzied emails came — “how are your DAUs looking today?” They were the same as the day before. Threads did contribute to the demise of T2/Pebble, but the effect wasn’t immediate. Rather, it came slowly, as we realized that every investor conversation we’d have in our fundraise would start and end with the fact that Threads is much bigger and much better resourced. Threads didn’t hurt directly our metrics, but it did negatively affect our fundraise. No additional Twitter alternatives have been launched since Threads, and I doubt that any will.
In Closing
I’m proud that we tried. We learned a lot. Our team has been a dream to work with. We built an supportive and caring small community.
Yes, we did lose most of the invested capital, and I feel like we have let our community down. But I think trying to build T2 / Pebble, and failing, was worth it.
Building consumer apps is hard. Consumers are fickle. Investors are worried they might embarrass themselves investing in the next big flameout. Growth sometimes stalls.
We’re not yet sure what we are doing next. Sarah, Dan, Mike, and I have loved working together on this. Thank you for letting us share these learnings with you.
P.S.: If this was interesting, we might write another episode on ourlearnings, please remember to follow me on Medium.
