Relay
RFI Coordinator
Routes, drafts, and chases RFIs. Finds precedent in past answers, specs, and drawings before anything goes out, and keeps ball-in-court honest.
Drafting a response package for RFI-142
AI agents
Fulcrum’s agents take on the legwork of construction: chasing RFIs, checking submittals, writing daily logs, reconciling invoices, and finding hazards. Each one has a job, a supervisor, and limits you set, and every step it takes is on the record.
How agents work
Agents in Fulcrum follow the same rules as your team, with a few stricter ones of their own.
Each has a role, a supervising person, a permission template, and an activity record, just like the people it works with.
Observe, Suggest, Act with approval, or Autonomous, set per agent and per tool.
Committing money, signing, changing permissions, and deleting records are never done by agents.
Every agent action records its reasoning, inputs, and tool calls, and shows up in the audit log.
The roster
Start from these role templates and name them whatever you like. These are the agents at work in Fulcrum’s sample workspace, and what each one is doing right now.
RFI Coordinator
Routes, drafts, and chases RFIs. Finds precedent in past answers, specs, and drawings before anything goes out, and keeps ball-in-court honest.
Drafting a response package for RFI-142
Submittal Reviewer
Checks submittals against spec sections and the drawings before they reach the design team. Flags every deviation with the paragraph it breaks.
Checking 23 73 13-01 AHU product data against spec
Document Controller
Ingests drawing and spec sets, splits and names sheets, slip-sheets revisions, clouds the deltas, and keeps everyone on the current set.
Watching for new drawing and spec uploads
Field Reporter
Builds the daily log from crew check-ins, gate deliveries, weather, photos, and voice notes. Takes meeting minutes and assigns the action items.
Compiling today’s daily log from 14 sources
Scheduler
Keeps the schedule honest. Reads daily logs and delivery data, forecasts slips, and models recovery options on the critical path.
Recovery options for the L24 pour delay are ready
Cost Controller
Reconciles invoices against schedules of values, prices change events, tracks commitments, and forecasts cost at completion.
Reconciling 14 subcontractor invoices against SOV
Safety Officer
Scans site photos and 360 captures for hazards, pre-fills JHAs, tracks certifications, and opens observations with the photo as evidence.
Reviewing 212 site photos from this morning’s walk
Quality & Punch
Turns walkthrough video into punch items with locations and trades, runs inspection checklists, and verifies closed items against new photos.
Next: L6 corridor walkthrough at 1:00 PM
Estimator & Buyout
Runs quantity takeoffs from drawings, levels bids, finds scope gaps between bidders, and drafts buyout recommendations.
Leveling 6 bids for package 05 12 00 Structural Steel
Owner Reporting
Assembles the monthly owner report: progress narrative, photos, cost and schedule summaries, and the risk register, led by the decisions the owner needs to make.
Paused while the owner’s new report format is set up
Role template
Tracks lead times, confirms deliveries, and flags long-lead risk.
Add it to any project and give it a name
Role template
Collects warranties, O&M manuals, and attic stock from every trade.
Add it to any project and give it a name
Need a role that isn’t here? Describe the job in plain words and Fulcrum drafts the agent for you.
Anatomy of a run
When an RFI comes back, Relay reads it, finds the location on the drawings, searches every past answer for precedent, and drafts a response. Sending it to the structural engineer is a message to another company, so it waits for Elena’s approval.
Relay Working
Drafting a response package for RFI-142
Autonomy
Turn the dial per agent and per tool. A new agent might start at Suggest, then earn Act with approval once its drafts are landing. Company guardrails always win, and agents can be made stricter, never looser.
The agent reads, notices, and reports. It never creates or changes a record.
Sentinel flags a missing guardrail in this morning’s photos and tells the Safety Manager.
The agent drafts the work and leaves it for a person to send, edit, or discard.
Vetta marks up a submittal with cited deviations and leaves the review to the project engineer.
The agent builds a plan and runs it the moment the right person approves it from their Inbox.
Relay packages the RFI-142 response and sends it to the engineer once Elena approves.
The agent acts on its own inside the tools, projects, and guardrails you set, and logs every step.
Atlas splits a 412-sheet drawing set, names every sheet, and clouds the deltas overnight.
Guardrails
Guardrails apply to every agent in the company. Admins set them once in AI governance; individual agents can be stricter, but never looser.
No agent, at any autonomy level
Agent proposes, a person approves
Inside the agent’s tools and projects
Agent studio
Write what you need, the way you would onboard a new hire: “Track long-lead equipment on Harbor Point and warn me if anything will arrive after it’s needed.” Fulcrum picks a role, writes the instructions, and suggests access, then you review it step by step.
Performance & spend
See actions taken, hours returned to the team, how often proposals are approved as written, and escalations to people. Compute budgets, reasoning tiers, and a company-wide cap keep spend predictable.
Fulcrum starts you with role templates for an RFI Coordinator, Submittal Reviewer, Document Controller, Field Reporter, Scheduler, Cost Controller, Safety Officer, Quality & Punch, Estimator & Buyout, Owner Reporting, Procurement Expediter, and Closeout Coordinator. You can also describe a job in plain words and Fulcrum drafts a new agent.
Every agent has a supervising person. The agent works inside that person’s access and can never exceed it, its proposals route to the right approver, and every action is attributed to the agent in the audit log.
Only with approval by default. Sending email, distributing RFIs, transmittals, and notices to other companies is an “ask first” action. You can pre-approve specific cases, such as reminders to the design team.
Each agent has a compute budget and reasoning tier (Deep, Balanced, or Fast). You set a monthly company cap and get alerts when any agent reaches 80% of its budget.
Yes. Any supervisor can pause their agent, and admins can pause every agent across the company with one switch. Running steps finish and nothing new starts until you resume.
We will show you Relay, Scribe, Ledger, and the rest working on projects like yours, and help you choose where to start.