I Made a Digital Clone of Myself. Here’s Why, and How, and How to Use and NOT USE Things Like This

There are now two of me.

One of me writes this blog, runs Markup AI, and still has to show up to meetings in person. The other lives in the cloud. Looks like me. Sounds like me. Never flubs a line, never needs a third take, never asks for a coffee break.

I built him with two tools — HeyGen for the video avatar, ElevenLabs for the voice clone. Total cost: $100. Total time saved: hundreds of hours and counting.

Here’s how it works, and what I’ve learned.

The Setup

The build is simpler than it sounds.

Step one: clone the voice. ElevenLabs wants 30 to 180 minutes of clean audio of you talking. More audio, cleaner audio, better clone. I did a mix of submitting audio I’d already recorded (e.g., podcasts and all-hands presentations) that were high quality, and some bespoke recording into their app with a great microphone.

Step two: build the avatar. HeyGen needs 3 to 5 minutes of video. Good lighting, decent framing, you talking like a person and not a hostage. I did a bunch of takes on this to get it right. Unscripted talking worked best.

Step three: connect the two via API – you get HeyGen to use the ElevenLabs voice instead of its own takes, which can take a little fiddling. I ended up re-recording the video sample several times once I saw how it was connecting to the audio via API to get it right without weird facial expressions.

Step four: upload a script, hit a button. You’re on video.

A weekend project. Not an engineering effort.

What I Can Actually Use It For

Three things, not ten:

Training videos. Anything instructional that used to mean blocking studio time — or re-recording every time the content changed. Now updating the video means updating the script.

Personalized welcome videos. A new customer can get me, on video, saying their name and speaking to their situation. People watch a video. They don’t read a templated email.

Quick updates. A walkthrough of a metric or a slide, for the team or a board member. Feels like I sent it. I did — I wrote every word. My avatar just handled the on-camera part.

The Caveat That Matters

Don’t overdo this. There’s a stiff authenticity price waiting to be paid.

The avatar works because the words are mine and the moments are ones where nobody expected me in the room anyway. A training video? Fine. A personalized welcome? Kind of delightful, actually. But the second people can’t tell whether they’re getting the real me or the synthetic me in a moment that matters, trust goes out the window. And trust is the whole game in business, and in life.

My rule: the avatar delivers my words. It never substitutes for my presence. Board meetings, hard conversations, anything where the human connection is the point — that’s analog me, every time. My secondary rule: depending on the use, I may choose to disclose that the video is the avatar. But TBH, if you have to disclose it too much or too fervently, you’re either using it too much or for the wrong use case.

The Bigger Point

The pattern holds across everything I’ve tried with AI: the win isn’t replacing judgment or relationships. It’s killing production friction. I was never going to record dozens of personalized videos a month. Now I can.

What’s Surreal

I can go from an idea and a few sentences, to my MattBot agent for script production, to the HeyGen/Eleven Labs avatar and finished product in seconds/minutes. And I can even have my EA do this for me. She and I have been joking about “needing to clone me” for years. Now we are uncomfortably close to doing just that, for at least some use cases.