An agent that does the work, inside the rules you set.
Most construction AI answers questions. These take the next step. They read your jobs, draft the change order, chase the RFI, and stop for a person on anything that costs money. Built on your process, in the system your crews already open.
The problem was never the AI. It was that nothing ever gave it permission to finish.
What it may do, per role
- Look up the job and its history
- Automatic
- Assign a contractor to the job
- Automatic
- Open an RFI
- Automatic
- Send a packet out for approval
- Needs a human
- Submit a pay application
- Needs a human
- Release a payment
- Needs a human
Scoped by role and enforced where the data lives, not by instructions in a prompt. Nothing typed into a chat box widens it.
Permission first, cleverness second
The reason most contractors have not put AI near a live job is that nobody could answer what happens when it is wrong. Three answers, built into the platform rather than written into a prompt.
Scoped to a role, not to a prompt
An agent inherits the permissions of the person it works for. A foreman's agent cannot reach the financial actions, because the foreman cannot. The scope is enforced where the data lives, so no instruction typed into a chat box can widen it.
A human signs anything that costs money
Every action in the registry is marked either automatic or always-human, and the ones that move money or go to a customer are always-human. Looking up a job or assigning a contractor runs on its own. Sending a packet for approval, submitting a pay application or releasing a payment waits in a queue for a person. That split ships with the platform today and is the same for every customer: changing it for your organization is part of an agent build, not a switch in your settings.
Every action leaves a row
Each tool the agent calls is recorded with what it did, which job it touched and why it chose that step. When a GC asks who approved a change in March, the answer is a query, not a memory.
Four steps, and the fourth is the one that matters
An agent that goes live on day one is a demo. An agent that spent three weeks proving it agrees with your best PM is a system.
Pick one decision, not a department
We start from a decision somebody makes the same way every week: which jobs slipped, which RFIs are past due, which close-outs are missing a photo. One decision is testable. 'Automate project management' is not.
Wire it to your records
The agent works on your jobs, your customers and your documents, in the platform, with your field names. No separate knowledge base to maintain and no export that goes stale the day after it runs.
Set the guardrails before the logic
We agree what it may do on its own, what needs a signature, and who it answers for. That gets written as permissions, not as instructions in a prompt, so it holds when the model changes.
Run it in shadow, then turn it on
It runs alongside your team first, choosing what it would have done while a person still does the work. You compare the two for a few weeks. The agent goes live on the day its answers stop surprising you, and not before.
What runs today, and what we are building
Worth being straight about, because the gap between an AI roadmap and an AI product is where most of this industry's money goes.
Live in the platform now
- The in-product assistant: ask for a job, a schedule or a customer in plain language and act on the answer
- Voice capture from the truck, so the job is logged before you drive away
- Role-scoped actions across jobs, field ops, documents, people, proposals and contracts
- An approval queue holding every action the registry marks as needing a person
- A logged row per action the assistant takes, per organization
The assistant is included on Professional and Enterprise. See what each plan includes.
In build, with design partners
- Schedule triage: what slipped overnight, who needs moving, and the three calls to make first
- RFI and submittal chasing, with the follow-up drafted and the clock tracked
- Change-order capture from a photo and a sentence spoken on site
- Close-out packets assembled and checked against the owner requirements
- Pay-application prep against the schedule of values, ready for a human to sign
We are building these with a small number of contractors on real jobs. If one of them is the decision eating your week, that is the conversation to have.
The platform, the agents, and the team that lands them
Most vendors sell one of these and leave you to find the other two. Pick the one that fits where you are.
ServiceIQ
Jobs, schedule, documents, approvals and billing in one system your crews actually open.
- One flat fee, not per seat, so growing the crew does not raise the bill
- Native HubSpot and QuickBooks, plus a branded portal for the customer
- Run a real job through it on day one, before you pay anything
Custom AI agents
Agents built on your process, scoped to a role, and stopped for a human wherever you say so.
- They act on your own records rather than answering questions about them
- An agent can never do what the person it works for cannot do
- Every action it takes is logged, and the risky ones wait for a sign-off
AI implementation
The team that builds the platform, putting AI into your operation and proving it moved a number.
- A readiness read and a roadmap before anybody writes code
- Integration and data work on the systems you already run
- Enablement for the crew, because adoption is where this usually dies
Bring the decision that eats your Mondays
Thirty minutes. We will tell you whether an agent is the right answer, and say so plainly when it is not.