The case at a glance: a $1,247 MRR case study in numbers
- MRR at month 4: $1,247
- Paying users: 43 (roughly $29/seat/month, my arithmetic — he never published a price)
- Timeline: built in May 2026, these numbers are September 2026
- Platforms supported: 1 (LinkedIn was considered and declined)
- Core suggestion engine, his own completeness estimate: 60%. Best feature: 30%
- Share of users sourced from his own X account: ~70%
He built ClimbX.so for himself, from a problem he could state in one sentence: he could write, he just didn't know what would get seen. The product learns from his users' own post stats, drafts in their tone, points at threads worth replying under, and schedules up to eight posts a day. He hasn't disclosed the stack. Everything here reconstructs to a boring one — Next.js, Postgres, one LLM API, platform API access, Stripe, call it $80-150/month at this size. No team, no calls, no client onboarding.
What he's actually selling
The product isn't the writer. The product is the hour of deciding.
Free tools generate fifty tweets in four seconds, which is exactly why generation is worth nothing. He shipped a scheduler that handles eight posts a day and not one user has asked for a ninth. What they ask, every morning, is what to post and where to comment. Output competes with ChatGPT. Judgment competes with nothing.
The single-platform refusal is the part most operators will misread as discipline. It's arithmetic. At 43 users, a learning loop split across two platforms has half the signal per platform and needs roughly double the users to make the same call with the same confidence. Staying on X is what lets a 43-user product behave like a 400-user one. Going multi-platform wouldn't have diluted his attention — it would have made the loop statistically mute.
And note where the money landed. The scheduler was the plan. The "what and where" was the throwaway. The throwaway is the line item.
Generation is free forever; judgment is the only line item left.