Four autonomous agents that run continuously in the background — monitoring your projects, surfacing risks, handling routine coordination, and making recommendations. No one has to ask them to act.
What "AI-Native" Means for BuildOS
Most software records what you tell it and shows it back to you. BuildOS goes further: it employs AI agents that watch what's happening across your projects, reason about what it means, and act — generating briefings, sending messages, creating alerts, and recommending decisions — without waiting for a user to trigger them.
These agents are not chatbots. They don't respond to questions. They run on schedules (daily, weekly) or in response to events (a delay reported, a permit issued, a task completed), and they write their outputs to the Notification Feed, to SMS, or to push notifications — wherever the right person will see them.
The Four Agents
Generates a personalized morning briefing for each superintendent — delivered to their mobile app before they arrive at the first site. The briefing compiles everything Mike needs to know for the day, synthesized from schedule data, weather forecasts, procurement alerts, and sub confirmation statuses across all of his active projects.
- Today's critical tasks — tasks on the critical path that are scheduled to start or continue today, across all projects
- Weather risk callout — any exterior tasks at risk from today's or tomorrow's forecast, with recommended actions
- Sub confirmation status — which subs have confirmed for upcoming tasks, which haven't responded
- Procurement alerts — any material deadlines within the next 14 days
- Yesterday's completions — tasks marked done by field workers since the last briefing, so Mike starts the day knowing where things stand
Monitors every material procurement item across every active project, every single day. It calculates each item's current status based on today's date and the must-order date, and generates feed cards when items cross into WARNING or CRITICAL territory. It also coordinates with the Tribunal when an item needs a purchasing decision.
Handles subcontractor communication automatically. It identifies tasks starting within 48–72 hours that need a subcontractor on-site, sends a structured SMS confirmation request, and interprets the response — without Mike having to make a single phone call. It understands natural language responses and can detect confirmations, delays, and partial progress updates.
| Sub Replies With | Agent Interprets As | Action Taken |
|---|---|---|
| "yes confirmed for Tuesday" | Confirmation | LOW priority feed card: "[Sub] confirmed for [task]" |
| "delayed, can't make it until Thursday" | Delay report | URGENT feed card + delay flag on task + CPM recalculation triggered |
| "60% done already" | Progress update | Task percent complete updated to 60% in the schedule |
| No reply after 24 hours | No response | URGENT feed card: "[Sub] has not confirmed — follow up required" |
The Tribunal is BuildOS's multi-model decision engine for procurement approvals. When a CRITICAL procurement item needs a purchasing decision, the Tribunal convenes three independent AI models, each with a distinct role. They vote independently and a consensus is reached — or the decision escalates to a human. The Tribunal's recommendation appears as an actionable feed card.
| Architect | Historian | Consensus | Result |
|---|---|---|---|
| YEA | YEA | Unanimous Approve | Recommendation: Approve order |
| NAY | NAY | Unanimous Reject | Recommendation: Do not order yet |
| YEA | NAY | Split — escalate | Coordinator breaks tie; notes disagreement |
| ABSTAIN (either) | — | Insufficient data | Escalate to human with context |
AI Autonomy Levels
BuildOS is designed around a model of graduated trust. The AI agents gain autonomy only as they demonstrate accuracy and as the organization explicitly opts into higher levels of automation. This is not just a design choice — it reflects a real principle: autonomous systems in business contexts should earn trust incrementally, with human oversight maintained until reliability is proven.
Agents surface insights, generate recommendations, and create alerts. Every consequential action requires a human to tap "Approve" in the feed. The Tribunal evaluates procurement decisions and presents a recommendation card — Tom or Sarah makes the final call with one tap.
For procurement decisions below a configurable dollar threshold (e.g., under $5,000), the Tribunal can auto-approve without a human tap — provided the item is clearly within budget and on schedule. Actions above the threshold still require human approval. The organization owner sets and can adjust the threshold at any time.
Fully autonomous procurement and coordination within approved parameters — no human approval required for any routine decision. Reserved for organizations that have demonstrated extensive track record with Level 2, and requires explicit approval from insurance carriers whose coverage would be affected by autonomous purchasing actions.
What Agents Will Never Do
Regardless of autonomy level, BuildOS agents operate within hard constraints. These are not configurable — they are built into the system:
| Action | Can an Agent Do This? |
|---|---|
| Delete or modify existing financial records | Never |
| Override a human's explicit rejection or dismissal | Never |
| Take legal actions (sign contracts, file permits) | Never |
| Access data outside the agent's assigned organization | Never |
| Send communications to clients or lenders | Never (subs only) |
| Approve its own recommendations | Never |
| Generate a morning briefing | Yes (always) |
| Send SMS to subcontractors | Yes (always) |
| Create feed cards | Yes (always) |
| Update task progress from SMS sub reports | Yes (always) |
| Recommend procurement approvals | Yes (Level 1+) |
| Auto-approve procurement below threshold | Level 2+ only (opt-in) |