Recently, while researching product cold starts, I ended up reading quite a bit about product competition. There were things I thought I understood before, but after reading, I realized I only knew about them—I didn't truly understand them.
The Old Formula Is Broken
In the past, the playbook for product competition was simple: whoever had the louder voice and more resources would win. Go big, plaster ads everywhere, and users would naturally come.
This playbook started to crumble in the AI space around early last year.
On January 20, 2025, DeepSeek released the R1 model. Almost nobody had heard of this company before—no brand recognition, no marketing budget. They just quietly posted the model to open-source communities and pushed a chat feature to their app.
Then came the Spring Festival.
In roughly seven days, DeepSeek reached over 100 million users. It topped the App Store charts in both China and the US, servers were overwhelmed, and they even suggested users try other AI assistants at one point. Yes, the servers couldn't handle the load—they were actually diverting traffic away. But DeepSeek didn't seem to care much; they just kept open-sourcing and updating.
The most representative comparison at the time was Doubao. Doubao is ByteDance's AI app, with arguably the largest resource investment domestically, and it had long held the top spot for DAU among AI apps. Before it, Kimi had reportedly spent hundreds of millions on marketing for growth, investing in Bilibili, Douyin, and the education market. Then when Doubao made its move, Kimi's buzz dropped significantly.
Then DeepSeek came along, spent not a single penny, and surpassed Doubao outright.
You thought those who spent big were safe, but then someone who spent nothing overtook you. That was quite a shock.
Blazing-Fast Iteration
Later, I went to check the version numbers for Claude Code and Codex, and I was a bit shocked.
Claude Code's latest version is 2.1.87. From its first release in February 2025 to now, they've shipped 365 versions total, averaging one every 1.1 days. You open it today, and there's likely something new.
Codex's latest stable version is 0.117.0. Since launching in April last year, they've released 132 stable versions, plus twenty to thirty alpha iterations per version—basically moving every day.
This isn't some small utility. Claude Code has over 10 million weekly npm downloads, Codex over 3 million. At this scale, iteration speed is measured in days.
Horizontal speed is even more absurd.
Last Thursday (March 27), DingTalk open-sourced workspace-cli, granularizing core features so AI agents could directly manipulate calendars, todos, and messages. The next day, Feishu open-sourced larksuite/cli, covering 11 business scenarios and over 200 commands. Another day later, WeChat Work's wecom-cli appeared on GitHub.
Three companies, three days, almost identical moves. Faced with a valid direction, follow-up speed is measured in days. How long can a "first-mover advantage" last in this environment? Maybe two to three weeks.
Doing Things That Don't Scale
While researching cold starts, I discovered a counterintuitive pattern: many products that later became huge didn't start by spending money, but rather by doing things that simply couldn't be scaled.
Paul Graham wrote "Do Things That Don't Scale" in 2013, which had a big impact in startup circles. The core idea is: don't think about scaling in the early days, use unscalable methods first.
Sounds like motivational fluff, but the case studies convince you otherwise.
Stripe's two founders, Patrick and John Collison—when someone said "I could try your payment product," they wouldn't send a link for you to figure out yourself. Instead, they'd say "Give me your laptop," and integrate Stripe into your code on the spot. Paul Graham thought this move was brilliant and named it the "Collison Installation," teaching it in every Y Combinator batch since. Airbnb did something similar. Early New York listings had terrible photos, taken casually on phones. Brian Chesky and Joe Gebbia flew over, rented a camera, and helped landlords shoot professional photos door-to-door. Listings with professional photos saw booking rates jump 2.5x immediately.
DoorDash was even more direct. The founders started as delivery drivers themselves, taking PDF menus from restaurants, using their own phone numbers as customer service. When someone ordered, they'd bike there and deliver it themselves. Pinterest's Ben Silbermann noticed early users were mostly design enthusiasts, so he went alone to offline meetups for design bloggers and recruited them one by one. Tinder took it further—Whitney Wolfe went to college sorority parties to promote it, with entry requiring the Tinder app installed on your phone.
The common thread in all these tactics: founders personally handled things, serving users one by one, doing things that couldn't scale. But because of this, that first batch of people genuinely felt "this thing is different," and then spontaneously spread the word for you.
Thinking about it, DeepSeek's path is essentially the same. Not through advertising, but because the product itself was so good others couldn't help but tell people.
A Two-Week Lead, Then Back to Zero
Putting all this together, my understanding of product competition has definitely changed.
Previously, gaps were created through resources and channels—whoever had more ads and stronger brands occupied more market. That advantage could last a long time. Now that's not working. Distribution infrastructure is too developed; something truly good can spread at almost zero cost. DeepSeek reached 100 million in seven days—the fastest ever. The supply side is completely different too; building a feature used to take months, now Claude Code ships daily, and DingTalk open-sources today with Feishu following tomorrow.
You do well, and your lead might last two to three weeks. If you don't keep getting better during those weeks, users will switch without hesitation. Claude Code and Codex are in this state now—you ship a feature, I match it, users switch between both.
The New Decisive Factor
So lately, I've become increasingly numb to "keeping an eye on competitors." I used to think this was basic product work: competitive analysis, differentiation positioning, finding white space. Now, in the AI space, spending too much time watching what others are doing is worse than spending time with your own users. The landscape you see today might change in two weeks.
I now think what really matters is whether the product itself works, whether users can feel "this is actually different." DeepSeek spent nothing on promotion, Claude Code barely does any acquisition, yet users are flooding in. Then there's the relationship with early users—how to find those who genuinely think this is cool and make them co-builders. Stripe's "Collison Installation" established this tight relationship between founders and users. In the early days, that relationship is worth far more than ten thousand users from ads.
The AI era does give product people more choices and freedom—you can focus more on the product itself rather than constantly competing over who has more resources and louder voices. Of course those things still matter, but they're no longer overwhelming.
At the end of the day, there's really just one question: Are you creating value, or just consuming others' attention?
Once you figure that out, a lot of second-guessing disappears.