Why I Failed to Make AGAR Work

I have written a lot of blogs about other people's failures. Goldmines almost losing everything to a hack. ZTE building exactly what its customers asked for and still watching nobody buy it. It's easier to write about someone else's mistake with clarity, because you're not the one who has to live inside it, you get to arrive after the fact, pick out the pattern, and hand it back as a lesson. This one is mine, and it's harder to write, because I still don't fully know if we were wrong or just early, and I don't have the comfort of distance yet to tell the difference. AGAR was supposed to be a place where the line between creator and viewer collapses, where AI generated content wasn't hiding, wasn't pretending to be something a person made, it was the point. That was the whole thesis, stated as plainly as I could state it in every pitch, every deck, every investor call. We went out to raise a seed round on that thesis. We did not close it. Here's what actually happened, as honestly as I can lay it out. The team problem, before the pitch even started The first wall we hit was structural, and it showed up before we'd even gotten deep into a pitch conversation. The product needed AI engineers, filmmakers, and people who genuinely understood both storytelling and model training sitting on the team or close to it, actually helping mature what the model could do, not just consuming an API and shipping a thin wrapper on top of someone else's foundation model. That's an expensive, slow team to build before you have revenue to justify carrying it, and most seed stage capital simply isn't structured to fund that kind of specialized team assembly before product market fit has been proven. In effect, I was asking investors to fund infrastructure and talent acquisition that looked much closer to a Series A ask, dressed up in a seed round's clothing, and that mismatch was visible on almost every single call, even when nobody said it out loud directly. Investors are pattern matching machines, and the pattern they were seeing didn't match the check size they were used to writing at this stage. The audience problem, which hurt more than any of the others The second wall was the one that actually hurt the most, because it wasn't a strategy mistake in the way the team sizing was, it was a timing mistake I didn't see coming with enough clarity, and by the time I saw it clearly, it was already shaping every conversation in the room before I'd said a word. I built for the Western, English speaking audience first, deliberately, because that's where the willingness to pay for digital content is real, the same argument I've made in almost every other blog on this site about why India specifically is hard to build for directly and monetize from. But the Western audience, right now, at this exact moment in time, hates AI content. Not dislikes, not is skeptical of, hates. The phrase AI slop isn't an internet joke anymore, it's become an instinctive, almost reflexive reaction, the moment someone suspects something was AI generated, trust collapses to zero before they've even properly looked at what's in front of them. I underestimated how fast that reaction had hardened into a reflex, and I underestimated how total it had become, and I built directly into the middle of it, at exactly the wrong moment for that particular bet. What frustrates me most, sitting with this months later, is that I don't think the underlying instinct behind AGAR was wrong. I wasn't trying to build something that replaces how people currently consume content on YouTube or Instagram, that was never the pitch, even if it sometimes got heard that way in a room. I was trying to build something closer to Wattpad, that specific, slightly embarrassing, deeply human appetite for cheesy fiction, for romance, for stories you'd never admit out loud to loving but will read at 1am on your phone anyway, under the blanket, telling yourself you'll stop after one more chapter. That appetite is real, and it's massive, Wattpad itself proved that at scale with entirely human written fiction, no AI involved anywhere in that story. I genuinely believe there's a version of that same appetite that would have welcomed AI generated stories the same way, if it had been positioned honestly as its own separate thing, not marketed as a replacement for anything, just a different room people could choose to walk into on purpose, the same way choosing to open Wattpad is already an active, informed choice about what kind of story you're about to get. But the phrase AI generated fiction platform, said cold, out loud, to almost anyone in 2025, reads instantly as exactly the thing they've already pre decided to distrust, before you've even had the chance to show them a single page of what you built. I never found the positioning that got past that reflex fast enough, before the raise ran out of runway to keep trying. The math that wasn't there yet The third wall was the coldest one of the three, because it wasn't about perception or timing at all, it was just the actual unit economics refusing to cooperate. Cost per generation versus revenue per generation did not make sense at the scale we needed to prove the model could eventually work as a real business. We would have been operating at a real, deep loss for longer than most seed investors are comfortable underwriting, and I did not have a clean enough answer yet for how that loss curve bent back toward something sustainable, what the exact mechanism was that would bring generation cost down or willingness to pay up, fast enough to matter. That's not a positioning problem, and it's not a market timing problem either, both of which at least have a story you can tell convincingly if you believe in it. That's just math that wasn't there yet, cold and unpersuadable, and no amount of conviction in a pitch deck fixes math that isn't there yet. The question I never quite landed And then there was the question that came up in almost every single investor conversation, in some form or another, sometimes gently, sometimes not. Why won't Google just build this. Why won't Meta. It's a fair question, genuinely fair, and I had an answer, I still believe the answer, but I don't think I always landed it in the room the way it deserved to land in the moment it mattered most. YouTube and Meta's platforms are built entirely on user generated content, their entire trust and safety architecture, their entire creator economy, their entire advertiser relationship, all of it is built around content coming from real, identifiable people. Retrofitting that into an AI native experience isn't a feature update for them, it's closer to rebuilding the foundation of a house while people are still living inside it, all while they're already dealing with a userbase that actively resents AI content showing up uninvited in feeds that were never supposed to contain it. AGAR was never trying to compete for that same audience, the people who feel ambushed by AI content appearing where they didn't ask for it. It was trying to build a room for the people who'd walk in on purpose, already knowing exactly what they were getting, the way someone opens Wattpad already knowing what kind of story they're about to read, choosing it rather than stumbling into it. I still believe that's a real, structurally different intent, and a real, different audience from the one Google or Meta would be risking by touching this inside their existing products. I just never got the runway to prove it at a scale large enough for an investor to believe it on faith, before the clock ran out on the round. Where this actually leaves me I don't have a clean bow to tie on this one, and I'm not going to force one just because a blog post usually wants a tidy ending. I don't think we were fundamentally wrong about where content is heading over the next decade. I think we were early on the audience's willingness to accept it, and exactly on time, in the worst possible way, on the economics not yet being solved, and that specific combination, being both early on sentiment and unsolved on cost at the same time, is one of the hardest possible places to raise money from, because there's no comparable company yet for an investor to point to as proof it works, and no clean, provable cost curve yet either to point to as proof it will eventually work. Investors weren't wrong to pass, given what was actually in front of them. I would probably pass too, sitting on their side of the table, looking at the same set of numbers I was looking at from mine. What I keep coming back to, more than anything else, is that AGAR not raising doesn't automatically mean the thesis was wrong. It might just mean the thesis was early, and early and wrong look identical from the outside until enough years pass that they stop looking identical. I don't know yet which one this turns out to be, and I'm not going to pretend I do for the sake of a cleaner story. I'm writing this down honestly, in full, rather than a polished, distanced version of it, because I'd rather have the real account on record while it's still raw, and because if the thesis does turn out to be right eventually, on some future attempt, mine or someone else's, I want to remember exactly where it actually broke this time, in precise detail, not a smoothed over, comfortable version of it written after enough time had passed to make it easier to tell.