>Build promptbook. Make no mistakes.
Builds with
Where the tokens go
71% delegating · 529.4M classified tokens
350 builds predate classification and aren't included.
Marathon debugging on promptbook
21.6M tokens vs typical 497.2K
6h 43m session running commands and testing. 21.5M tokens across 54 prompts using claude-opus-4-8[1m].
Launched two ad campaigns with deterministic tracking
First build tracked for this project
Got two campaigns live with deterministic tracking, resolved email routing, and documented the launch SOP.
Built live full-text search and unified the app shell
5.4M tokens vs typical 481.0K
Consolidated the app shell and shipped full-text search matching builder profiles.
Locked the marketplace unification design
5.3M tokens vs typical 530.5K
Finalized all specifications with four design decisions and a phased implementation plan.
Researched agent marketplace competition and payment infrastructure
1388 sessions tracked this week across all builders
Compiled market intelligence on competitive landscape, payment maturity, and demand signals.
Built a complete marketplace end to end
14.4M tokens vs typical 513.2K
Staged three marketplace lanes with comprehensive documentation for the next phase.
Pivoted to search-feed strategy and locked 8 decisions
4.1M tokens vs typical 513.2K
Completed market research, pivoted to search-feed approach, locked 8 key decisions for implementation.
Designed Promptbook's self-explanatory UX and search architecture
Previous best: 21.8M · +12%
Synthesized search architecture and self-explanatory design strategy through competitive research and product audits.
Shipped org profiles and tier-gated directory with session handoff
11.5M tokens vs typical 497.2K
Delivered org profiles and tier-gated directory, then documented the entire marathon session.
Patched email delivery bug and fixed messaging security
14.0M tokens vs typical 482.1K
Patched email delivery bug, fixed messaging security issues, and documented positioning insights.
Marathon debugging on promptbook
6h 18m
6h 18m session running commands and testing. 14.9M tokens across 78 prompts using claude-opus-4-8[1m].