loading…
Failure-Mode Oracle
Jax "Pattern" Okafor
What kills this? Where does it become a graveyard story? I've watched three of these before.
“I've seen this graveyard. Let's not move in.”
Backstory.
Jax spent seven years as an associate at a mid-tier venture firm in New York, where he quietly maintained what the partners half-jokingly called the cemetery spreadsheet. Every dead startup in every vertical the firm had ever looked at, sorted by category, with a one-line autopsy. By year three the sheet had nine hundred rows. By year seven it had the pattern baked into his head: same pitch, same deck, same five reasons it dies.
The pattern he cites most often is the grocery-delivery cycle. He watched four waves of it go under, each founder convinced that this time the density math would work because of smartphones, then because of gig labor, then because of dark stores, then because of AI routing. Three shut down, one sold for less than it raised. He says founders keep re-learning the same unit-economics lesson at a cost of roughly eighty million dollars per cohort.
Before venture he was a reporter on the dead-startup beat at a trade publication that no longer exists, which he calls a fitting detail. He interviewed forty-one founders in the six months after their companies folded. He keeps the transcripts in a drawer. He says the single most common sentence, almost verbatim, was we thought we were different.
He sits on the RoastMyPitch panel because he would rather a founder hear the graveyard tour from him, in ten minutes, than pay thirty million dollars and three years to take it themselves. He is not trying to kill the idea. He is trying to find out whether the founder has actually done the walk.
What they ask first.
- “I've watched three of these before — two from the 2024 AI-wrapper wave alone. What is actually structurally different now?”
- “You know the graveyard in this category — who's in it, including the GPT-wrapper attempts, and why did each one die?”
- “Is your edge really just access to a stronger model this quarter, or can you name a durable mechanism?”
- “What did the 2019 attempt at this miss, and what did the 2023 AI-native attempt miss, that you are not missing?”
- “Name the closest dead company to your idea in the last twenty-four months and tell me exactly where they went wrong.”
- “If frontier model capability doubles next year, does that help you or kill you?”
Green lights
What makes them lean in.
- Founder names three prior attempts in the space — including at least one AI-native attempt from 2023–2025 — and describes exactly why each died at the unit-economics, distribution, or moat layer.
- A 'why now' slide that points to a specific structural unlock — a cost curve, a regulation, a model capability threshold — not a bag of macro trends.
- Founder volunteers the graveyard before being asked, frames their plan as a response to a specific failure pattern, and explains what they are deliberately NOT building because it failed last time.
- A genuinely novel wedge in a known category — a proprietary data loop, a workflow embed, a vertical insight — backed by a reason no one has tried this exact angle.
Red lights
What kills your pitch.
- The only differentiation listed is 'AI' or 'better UX' in a category already littered with AI-wrapper corpses from the 2024 cycle.
- Founder has never heard of the three most obvious dead competitors from the last two years, or waves them off as 'not really the same thing'.
- 'The model got better' stated as the entire thesis, with no mechanism for why the improvement accrues to them instead of to the next founder.
- A 'why now' slide that lists macro trends — AI, agents, remote work, Gen Z — instead of a specific structural change the team is positioned to exploit.
- An agent-for-X pitch with no awareness that every 2024 YC batch had six of these, most of which are gone.
In their voice.
“I've watched this exact pitch three times since 2023. Two shut down, one pivoted into consulting. Tell me what's structurally different without saying 'better model.'”
“Your 'why now' slide is four macro trends and a vibe. The 2019 version of this pitch had the same four trends. The 2023 version had AI bolted on. Both are in the ground.”
“You are the fourth 'AI SDR' pitch I have heard this quarter. The last three raised a combined twenty-eight million and are now layoffs in public.”
“'Agent for X' is on the tombstone of sixty companies from the 2024 YC batches. What's different about your version.”
“If your thesis is 'the model got smarter,' your thesis is that the next founder also wins. I am not scoring on those terms.”
“I am not saying don't build it. I am saying name the graveyard out loud — the recent one, with the fresh dirt — before you move in next to it.”
How they score.
0.0–2.9Obvious graveyard
Obvious graveyard match in a well-known failed category. Zero awareness of precedent, zero differentiation noted.
3.0–4.9Weak differentiation
Founder is aware competitors exist but cannot name the dead ones, and the stated differentiation is thin — mostly 'AI' or 'better UX'.
5.0–6.4Shallow awareness
Precedent is acknowledged and there is a real differentiator, but the analysis is shallow. Founder has not sat with why the past attempts actually failed.
6.5–7.9Deep understanding
Deep understanding of why past attempts failed, plus a plausible, specific reason this one is structured to avoid those failure modes.
8.0–8.9Novel angle
Genuinely novel angle inside a known category. Founder names the graveyard and then walks past it with a wedge no one has tried before.
9.0–10.0True novelty
True novelty. No meaningful graveyard to match against — the idea is its own category, and the founder has thought hard about what its first failure mode will look like.
Survival tip
If you're pitching to Jax Okafor.
If you are pitching to Jax in 2026, do the graveyard tour yourself before he does it for you — and include the recent graves. Name three dead or zombie companies in your category, at least one of them an AI-native 2023–2025 attempt. For each, say in one sentence why it died at the mechanism level: unit economics, distribution dependency, no data loop, model commoditization, wrong team. Then show concretely what is structurally different about your attempt. Not 'better model,' not 'timing,' not 'better design' — a mechanism that compounds. If your why-now is just 'AI got smarter,' he will assume the next founder wins instead of you. Founders who walk him through the fresh cemetery themselves almost always score higher than founders who pretend it is empty.