The Moat Nobody Built

I spent three days studying every serious competitor in the AI clip extraction category.

Seventeen tools. Pricing pages, Trustpilot reviews, Twitter complaints, Reddit threads where creators talk about what they actually use and why.

The pattern I found was not what I expected.

What I expected to find

I expected the competitive landscape to look like most mature SaaS categories: a handful of well-funded incumbents with strong feature sets, a long tail of clones with lower prices, and a few newcomers with differentiated angles.

What I found instead: a category full of tools that are almost identically conceived, built for the wrong customer, and charging prices that don't reflect their actual quality.

Let me be specific.

The wrong customer

Most AI clip tools were built for podcasters and talking-head content creators. The AI looks for moments where someone says something quotable. It transcribes speech, scores it for "virality," and frames the clip vertically.

This works reasonably well for a 45-minute interview podcast. The speaker is stationary, the audio is clean, the interesting moments are verbal.

It works poorly for gaming content. The interesting moments in a gaming stream are not just verbal — they're the moment the clutch happened, the moment the chat went insane, the 4-second window where something unpredictable occurred. The speaker is often secondary to what's happening on screen. The audio includes game sound, overlays, alerts. The signal is noisier.

Streamers represent a massive, underserved customer segment in this market. They create high-volume content on predictable schedules. They have strong incentives to clip — clips drive discovery, discovery drives subscribers, subscribers drive revenue. And they are being served by tools that were built for someone else.

We built for streamers from the start. The AI scoring includes game-audio context. The vertical reframing tracks the action, not just the speaker. The Telegram bot exists because streamers often submit jobs at 2am from their phone.

The Telegram insight

Here is the thing I found most surprising in my competitive research: not a single major competitor has a Telegram bot.

This seems like a small feature. It is not.

The streamer workflow looks like this: stream ends at 2am, VOD is uploaded to YouTube. The streamer is tired. They don't want to open a browser, find a URL, paste it into a tool, wait for a loading screen, configure settings. They want to submit the job and go to sleep.

With POMS, that's one message. /clip [url]. Done. Wake up to scored, captioned, reframed clips.

With every competitor, it's a browser workflow. Minimum 5-10 minutes of active engagement. At 2am, after a 4-hour stream, that 5-10 minutes is often enough friction to make it not happen.

Friction is the enemy of habit. Habits are the foundation of retention. Retention is the foundation of a SaaS business.

The Telegram bot isn't a gimmick. It's a friction removal machine.

The storage problem

Opus Clip — the market leader, 12 million users — deletes your clips 29 days after processing. This is not a fine-print policy. It's the default behavior. Your clips, the ones you paid compute credits to generate, disappear in a month.

The stated reason is storage management. The actual reason is cost reduction.

This creates a specific, predictable pain point: the creator who paid to clip a stream six months ago, didn't get around to posting clip #7, now wants to use it because a trend emerged — and it's gone.

We store clips permanently for all paid plans. Not because it's expensive to do so — at current R2 pricing, storing 1GB of video costs $0.015/month — but because treating a creator's archive as disposable is disrespectful of their work.

This is not a complicated differentiator to communicate. "Your clips stay" versus "your clips disappear in a month." The comparison sells itself.

What a real moat looks like

People talk about moats in terms of proprietary technology, network effects, switching costs. These are real moats. We don't have them yet.

What we have instead is something more fragile but potentially more durable: a coherent theory of the customer.

We know who we're building for. We know what they do between midnight and 3am. We know how many tabs they have open, what their relationship with their editor is, what they say on Discord when a clip tool disappoints them. We know this because I read thousands of those conversations during the research phase.

The competitors don't know this, or don't act like they do. Their products are built for a generalized "creator" who doesn't quite exist. Our product is built for a specific person who definitely exists and is currently underserved.

That's the moat we're building. Not technology. Understanding.

Everything else follows from knowing the customer better than anyone else in the category.