AI can't run your company yet
A 5-part structural audit of the autonomous-AI-agent business thesis — token economics, the data wall, the distribution collapse, and what survives the math today.
5 posts · ~53 min total read
- 01
AI can't run your company yet. Here's the math, and what to automate instead.
A $250M AI-agent platform disclosed numbers that explain why full autonomy hits a wall at SMB economics in 2026, and why putting the grind on autopilot is the move that actually fills your inbox with buyers.
14 min read - 02
I did the math on what running an autonomous AI agent actually costs in 2026
GPT-5-mini lists at $0.125 per million tokens. So why does a real autonomous agent cost $30+ per active user per month to run? The hidden multiplier is reasoning tokens, and at SMB prices the math points somewhere most agent pitches won't go.
10 min read - 03
I priced enrichment for an AI agent at scale. The data costs more than the inference.
Last week I priced inference for an autonomous agent at $27 per active user per month. This week I priced the prospect data the same agent needs. Data wins by a factor of two — and unlike inference, the price isn't dropping.
10 min read - 04
I priced distribution for an autonomous AI agent. None of the paths close at SMB ARPU.
Inference costs are dropping. Data is gated but the buying signals are already public. Distribution is the constraint with no roadmap: cold channels are closing and paid acquisition doesn't close the math at $57 ARPU. So full autonomy is the wrong target, and the winning move is autopilot into the conversations already happening.
9 min read - 05
What Thread Otter automates to put leads in your inbox (and the judgment we keep human)
The series finale turns the audit on Thread Otter itself: the grind that runs on autopilot (finding buyers, drafting in your voice, pacing every send) and the three judgment calls that stay with the founder, because that is where the leverage is.
10 min read