Build record · tools
AI Build Cost Workbench
Live · v2, August 2026
Custom AI projects get quoted as one number, even though discovery, research, architecture, implementation, testing, documentation, launch and maintenance are different kinds and amounts of work. That opacity hurts the buyer and the builder equally.
Controlled visual demonstration
Current live interface
This sandboxed preview shows the current public interface. Open the full site to use the build in its intended window.
Verified build record
What is actually running, and what it is not.
LiveLive · v2, August 2026
The problem. Custom AI projects get quoted as one number, even though discovery, research, architecture, implementation, testing, documentation, launch and maintenance are different kinds and amounts of work. That opacity hurts the buyer and the builder equally.
What it does. Walks six stages — project brief, editable labour architecture across eight workstreams, model and API workloads, infrastructure, one-time and recurring costs, named uncertainty drivers — and returns cost composition, sensitivity drivers, build and monthly ranges, and a first-year total. Scenarios save locally, share by URL, print to PDF, and export as CSV and JSON. The model-price catalogue is dated and editable rather than hidden behind a project-type selector.
Built on. A static browser application with local persistence and URL-encoded scenario sharing, served from Vercel.
What it does not establish. Outputs are the user's own planning scenarios. They are not quotes, and they are not independently validated market benchmarks.
This record is generated from the same inventory as the complete build index: the production URL was checked at the last regeneration, and the status line above repeats what the build reports about itself rather than restating an intention.