Deterministic Structure for Probabilistic Intelligence: SAIL, Our Answer to Vibe Coding
Year: 2026
Type: preprint
Abstract: SAIL is a methodology in which AI agents build and maintain software continuously. The agents work from a chain-linked map of the product, through staged iteration, scripted review and self-correcting measurement. To our knowledge, SAIL is the first methodology of its kind that has AI agents build and maintain software continuously from a chain-linked map of the product, through staged iteration, scripted review and self-correcting measurement. The paper reports an 88-day measured record of SAIL on a client product as the evidence that it runs. All of the product's work was done by AI agents: 2,960 of 2,960 units, with no product code written by a person. Over 88 days, 52% of working hours had no human input at all; in the rest, the person's input went to the process itself and to directing and approving the product, not to writing it. 44% of finished work completed overnight, and idle time with work waiting fell to 2% in the final week (10 to 16 November 2026). The product stood at its measured size 62 days after its first map record, against 470 to 1,410 days from the published COCOMO II schedule equation for that size; a private-beta site was live after 48 days.
Author: Vasquez, Robert
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7234727330
