Why I built Regent (and what every social tool keeps getting wrong)
I'm Sumit. I'm the founder of heyregent. This is the story of why I'm building it.
It's not the story I'd tell an investor. It's the story I'd tell another creator who was about to make the same expensive mistakes I made.
#The first attempt cost more than my rent
I wanted to start posting on social — reels, mostly. Talking-head stuff, sharing what I was learning. The problem: I hate filming. I hate setting up the camera. I hate the lighting. I hate doing 8 takes of the same line because I flubbed it. I'd rather build than perform.
So I thought: AI avatar. Let me clone myself, generate the reels, ship them while I work on the actual product.
Three months and a stack of subscriptions later, I had:
- An AI avatar tool that charged per minute of rendered video, and the output looked like a corporate stock photo come to life. $80/month. Couldn't use it.
- A voice cloning tool that needed 30 minutes of training audio (I gave it 2 hours to be safe) and produced a voice that sounded like me read by a hostage. $30/month. Couldn't ship it.
- A scheduler that charged per platform and didn't even handle Instagram Reels natively. $25/month. Worked, but only for half the platforms I needed.
- A captioning tool because the scheduler couldn't auto-caption. $15/month.
- A competitor research workflow that was just me opening Instagram in incognito and taking screenshots into Notion. Free. Worth less than free.
That's $150/month, four browser tabs, three different file format exports, and a workflow that took 6 hours per reel.
And the reels I actually shipped didn't perform. The avatar was uncanny. The voice was off. The schedule was wrong because I was guessing peak times. The hooks were generic because I wasn't actually studying competitors, I was guessing at what would work.
I shipped 4 reels in three months. Spent ~$500 in subscriptions. Got 200 followers. Quit.
#The realization
The tools weren't bad individually. The voice clone was 85% there. The avatar was 70% there. The scheduler did its job.
The problem was that nobody had built the whole loop. Each tool solved one piece, and the joints between them — the file format conversions, the manual handoffs, the "now copy the avatar video into the scheduler, but wait, the scheduler doesn't accept this codec" — added 80% of the friction.
I'm a builder. I should be building. Instead I was a janitor between four AI tools, paying for the privilege.
The realization: nobody was going to build the end-to-end thing. Every existing tool was a venture-backed company optimizing for their slice. They had no incentive to absorb the slices next to them. The integration problem would stay a problem.
Unless someone just did it.
#What Regent is
Regent is the loop I wanted. Built so I'd never have to be a janitor between AI tools again.
You give it three competitor handles. It watches them daily and extracts the hook patterns from their best posts. (Regent Insight.)
You upload one photo. It trains an avatar that holds a 60-second monologue without breaking. (Regent Avatar.)
You read 15 seconds into your phone. It clones your voice at 95% accuracy. (Regent Voice Engine.)
You hit "generate this week." It writes scripts using the patterns from your competitors, in your tone, on topics it picked from the gaps in your competitors' coverage. (Regent Calendar.)
You approve the topics. It renders the reels in 4K with your face and voice. Captions in your tone. Posted at peak time across Instagram, TikTok, X, YouTube Shorts, LinkedIn. (Regent Publish.)
DMs from commenters get qualified, routed, and replied to in your voice — booking calls into your calendar while you sleep. (Regent Funnels.)
One stack. One workflow. One subscription. No file format conversions. No tab-switching. No janitor work.
#What social tools keep getting wrong
After three months of stitching tools together, I have hot takes. The tools that exist today suffer from one of three diseases:
1. They solve one slice and charge full SaaS pricing. $30/month for voice cloning. $80/month for avatars. $40/month for scheduling. You end up paying $150–250/month for one creator's worth of output. The math doesn't work for a beginner. The math barely works for a creator with 50k followers.
2. They optimize for the wrong audience. Avatar tools target corporate training videos, not personal-brand reels. Voice tools target audiobook narration, not casual creator monologues. Scheduling tools target social media managers running 12 brands, not one person running their own. The output reflects the audience — corporate, formal, designed for compliance. Not the casual, fast, you-feeling vibe that wins on reels.
3. They require a video editor. Even when the AI generates the raw clip, you still have to load it into CapCut to add captions, fix timing, add B-roll cuts. The very thing you were trying to skip — opening an editor — is still in the workflow.
Regent attacks all three. One subscription. Built for personal creators specifically. Zero video editor in the loop.
#What happens if Regent works
If Regent ships what I'm building, the math changes:
- A creator at 0 followers can ship 30 reels a month for less than the cost of one freelance editor.
- A founder running a SaaS can show up on social without the time tax.
- A team-of-one indie hacker can build in public without sacrificing the build time.
- A creator at 500k followers can scale their output 3x without hiring an editor or assistant.
That's the bet.
#What happens if it doesn't
If Regent doesn't work, I go back to the world I came from — building products that aren't this, working with founders who aren't me, living in the corner of the internet where I was before.
I'd rather not.
So I'm building it. Loud, in public, with the first 100 creators shaping the roadmap directly.
If you want in — the Creator Beta is open. 100 spots. Founders rate. Direct line to me on Discord. Apply, and if we're a fit, you'll be shipping reels in your face and voice within 24 hours of approval.
That's the loop I wish I had three months ago. I'm building it now so you don't have to.
— Sumit
Read next:
