AI can build a narrow prototype quickly, but there is no reliable universal time for a full product. The schedule depends on unresolved product decisions, workflow count, integrations, data migration, security, testing, deployment, and the level of reliability users expect. Generated code is only one part of delivery.
Estimate by vertical slices rather than pages or prompts. List the essential user journeys and classify each as known, uncertain, or high risk. For every slice, include requirements, design, implementation, integration, verification, review, deployment, and repair. Build the riskiest thin slice first; it reveals whether authentication, payments, external APIs, or data rules invalidate the plan. Track accepted result time and review effort. Use the first completed slices to update the forecast instead of extrapolating from a polished demo.
Define what “full product” means. A private experiment, paid beta, internal system, and regulated public service have different readiness gates. Include monitoring, backups, support, incident response, privacy, accessibility, and maintenance where the context requires them.
If the goal is speed, reduce scope before reducing checks. Choose one user, one painful job, and one measurable outcome. AI can compress many implementation steps, but the honest release date comes from verified slices and remaining risk. Reforecast after each milestone rather than promising a fixed build time at the start.
