What's being built, in what order, and why โ from the initial MVP through full autonomous operation. Three phases spanning 12+ months.
Phase 1 โ MVP Sprints 1โ8 ยท 16 weeks
The MVP delivers the six capabilities that eliminate the most administrative overhead and provide immediate, demonstrable value to a GC firm. The sprint cadence is two weeks. The MVP is designed to be shippable โ a complete, stable product that a real firm could run their operations on โ not a prototype.
Real-time portfolio financial view for Tom: estimated vs. committed vs. actual across all active projects, with per-phase drill-down, AR aging buckets, and invoice approval workflow for Sarah. Replaces the monthly 4-hour spreadsheet reconciliation with an 8-minute dashboard review.
Port and harden the battle-tested CPM physics engine from the legacy system. Preserve the deterministic integer math, all four dependency types, and the golden master test suite. Add the web-based Gantt chart and schedule recalculation API. This is the scheduling backbone that every other feature depends on.
Integrate Tomorrow.io hourly weather forecasts into the schedule duration model. Weather-sensitive tasks (WBS 1โ9 and exterior 13.x) receive adjusted durations based on actual site-specific forecasts โ not generic multipliers. Proactive weather alerts in the Daily Briefing and Feed for at-risk exterior tasks.
Launch the Daily Focus Agent and Flutter mobile app simultaneously. Mike gets his AI-generated morning briefing via push notification every day before 7 AM. Carlos gets the field app for progress reporting, GPS check-ins, and daily field logs โ with full offline capability and bilingual (EN/ES) support.
Launch the Procurement Agent's daily monitoring loop and the Sub Liaison Agent's SMS confirmation system. Add the Tribunal ConsensusEngine at Level 1 (recommend-only). Sarah sees procurement deadlines automatically โ the Tribunal evaluates critical items and surfaces recommendations for Tom to approve with one tap.
Built in parallel with M3. The full Kanban pipeline CRM: six stages, estimate management, permit tracking, pipeline analytics with weighted revenue forecast, and the atomic Permit Issuance Gate that automatically creates the construction project and initializes the schedule at the moment a permit is approved.
Phase 2 โ Post-MVP Months 5โ10
Phase 2 deepens the platform's capabilities, expands automation, and adds the operational management modules (fleet, HR) that complete the full GC back-office picture.
Full equipment tracking: asset registration, date-range allocations with database-enforced conflict prevention, maintenance reminder automation, and the fleet dashboard showing where every piece of equipment is at any given time. Eliminates the whiteboard and phone-tag system for equipment coordination.
Employee records with digital certification management and automatic expiry alerting. Contractor licenses, OSHA certificates, and insurance certificates โ all tracked with 30-day advance warnings and Critical-priority feed cards for anything expired. Replaces the binder and the calendar reminder system.
Automatic generation of AIA G702 (Application for Payment) and G703 (Schedule of Values) forms using data already in BuildOS โ WBS phase budgets, CPM percent complete, and previously certified amounts. Saves 2โ4 hours per project per draw cycle and eliminates manual assembly from spreadsheets.
Crew and equipment availability integrated directly into the CPM schedule. Tasks that require specific equipment will surface warnings if the needed asset is not allocated during that window. The schedule optimizer will suggest allocation adjustments to avoid equipment conflicts before they cause delays. Mike can plan crew assignments with confidence that the schedule reflects real resource availability.
After the Tribunal has accumulated 100+ evaluated procurement decisions with >90% accuracy vs. human decisions, Level 2 autonomy unlocks as an opt-in feature. Procurement decisions below a configurable dollar threshold are auto-approved without a human tap. Tom sets the threshold. Decisions above it still require his approval. The Tribunal accuracy rate is reported continuously.
Phase 3 โ Future Horizons 12+ months
Phase 3 represents the long-term vision for BuildOS โ where the platform moves from a powerful management tool to a partially autonomous operating partner. Each of these capabilities requires significant trust-building with users, regulatory consideration, or market maturity before they can be responsibly deployed.
All routine procurement decisions executed autonomously, within approved parameters, with no human approval required. Requires insurance carrier approval, legal review of autonomous action scope, and owner sign-off on a liability framework. Not available until Level 2 has demonstrated sustained accuracy over an extended period.
A read-only portal for construction lenders and project owners to view real-time schedule progress, financial status, and draw request documentation โ without requiring a BuildOS account. Eliminates the monthly PDF report and the lender phone call.
A lightweight web portal for subcontractors to view their confirmed tasks, submit progress updates, and upload their own invoices and certificates of insurance โ without the GC needing to enter the data manually. Two-way integration replaces the SMS-only coordination model.
Full CAD currency support throughout the platform (already architected โ USD/CAD separation is in the data model from day one), plus Canadian regulatory adaptations (building codes, permit processes, HST/GST handling) and French-language support for Quebec market firms.
An AI model trained on BuildOS project history that predicts schedule slip probability for tasks before the delay occurs โ based on patterns like sub responsiveness rates, weather exposure, inspection history, and crew velocity on similar tasks. Surfaces leading indicators, not just lagging ones.
The Autonomy Evolution
The roadmap has a consistent through-line: BuildOS earns autonomy incrementally, through demonstrated accuracy and explicit trust grants from the organization. This is not a marketing choice โ it reflects a fundamental design principle about how AI systems should operate in business contexts where errors have real financial and legal consequences.
| Phase | Autonomy Level | Human Role | Gate to Advance |
|---|---|---|---|
| MVP | Level 1 โ Recommend only | Approves every action. Agents surface and recommend; humans decide. | โ |
| Phase 2 | Level 2 โ Auto-approve within threshold | Sets the threshold. Reviews edge cases. Receives accuracy reports. | >90% accuracy over 100+ decisions + owner opt-in |
| Phase 3 | Level 3 โ Fully autonomous | Sets parameters. Reviews exceptions. Oversees at policy level. | Extended Level 2 track record + insurance carrier approval + legal review |