Blue VariablesAI agents, deployed inside capital-project workflows.
Pick the workflow that's burning your team's time — submittals, RFIs, daily reports, lookaheads. In three weeks you'll have a working agent running on your own documents, a measured accuracy number, and an honest answer on whether to run it.
For owners, EPCs, contractors, and the technology vendors who serve them.
Founding clients: three organizations this fall get the working session free.
Domain depth and shipped systems, in the same room.
Owners, EPCs and contractors bring us a workflow; we bring the people who have run it and the systems to automate it.
We've run the workflow.
15 years in capital projects — Construction Tech / Industrial Construction / CII / AWP / Lean / project controls. PhD in construction engineering and project management. We know what an IWP, a lookahead and a submittal log are before the kickoff call.
We build and run it — not just advise.
Not a slide deck, not a vendor demo. A working agent on your documents, every step traced, evaluated against your past cases, improving with use. You keep the workflow, the prompts, the evals — and you can export all of it.
We're vendor-neutral.
Forty-plus AI products exist for AEC. Often the right answer is to configure what you already own. We'll tell you which, and what still needs an agent.
Three steps. One workflow at a time.
Working session — 90 minutes
We map one workflow with your team and run the agent on a sample of your real documents, live. Two-page readout within three days: what it did, where it failed, whether a sprint is worth it.
Free for founding clients.
AI Workflow Sprint — 15 business days
Process map → working agent on your data → evaluation on 50–200 real past cases → build / buy / configure / wait decision with the cost equation → handover with 60 days of platform access.
About six hours of your team's time. Months two and three are the handover window: your team keeps using the workflow while IT signs off and the production deployment is set up.
Run it — from month four
Once it's in production, your team runs the workflow with our success team alongside for two quarters — integrations, training, monitoring.
Tapering to zero by month nine.
A Fractional AI Lead, two days a month.
An experienced, vendor-neutral operator in your leadership meetings — someone who has run these workflows and shipped the systems behind them. Sorting the forty-plus AEC AI products, deciding what to configure, buy, build or wait on, and setting up the evaluation discipline before you commit to any of it.
- Monthly AI roadmap and vendor calls, with the reasoning written down
- Pilot and proposal reviews against measured accuracy, not demos
- A standing answer to "what should we do about AI" for the board and the program
- Converts to a sprint whenever a workflow is ready — no re-selling
Workflows we're sprint-ready for
Document-heavy, repetitive, expensive when they slip. Bring one of these — or the one we haven't listed.
RFI and submittal triage
First-draft responses with spec paragraph and package page cited. Critical categories routed to a person by rule.
Spec and drawing review
Requirements extracted into a checklist and compared item by item — meets, does not meet, not addressed, needs human.
Daily / safety / quality reports
Field reports rolled into a weekly owner narrative that reads like your best PM wrote it.
Lookahead constraint flagging
Constraints surfaced from P6 and procurement data before they hit the three-week window.
Change-order package assembly
Backup, narrative and pricing pulled together in your house format from the source records.
IWP generation for AWP programs
Installation work packages assembled from CWPs, model data and constraint status.
Vendor document reconciliation
Vendor submissions checked against the document register and the purchase order.
The one we haven't listed
If it's document-heavy and repeats every week, it's probably a good sprint. Tell us about it.
Bring your workflowThree organizations this fall. Structured so you don't lose if it doesn't work.
A free working session on one workflow. Then, if it's worth it, the sprint at a reduced founding-client fee with a written satisfaction clause: if at the readout you judge the sprint didn't deliver what the SOW promised, you pay only the first half and keep everything.
In return: your honest feedback, a reference if we earn it, and the right to describe the work anonymously.
Reduced for the first three clients, with the satisfaction clause in writing. 15 business days · one workflow · working agent on your data · eval on real cases · decision memo · 60 days of platform access.
Yours: the workflow, the prompts, the evals, the data. Ours: the platform. Both in writing.
Workflow definitions, prompts, rules, evaluation sets and outputs are assigned to you. The platform that executes them, and its code, are licensed — not transferred.
The export bundle is one command and works in every license state — including after expiry. Prompts, rules, evaluation sets and SOPs travel to any runtime; the README explains the rest.
Your Azure, AWS or GCP subscription from a template your team applies; the AI platform you already license — Azure AI Foundry, Copilot Studio, Amazon Bedrock, Google AI Studio, OpenAI, or Claude; or our hosted environment with per-tenant keys. You pick the boundary once, in the SOW. Data stays inside it; the only thing that crosses is metadata telemetry, if you turn it on, visible byte for byte in the audit log.
"I spent a decade inside these workflows. Now I build the systems the agents run on."
Under pressure to "do AI" on a program?
Book a callPick one workflow. We'll tell you in 30 minutes whether it's a good first sprint — and show you the agent on your kind of documents.
Not ready to build? Ask about a Fractional AI Lead instead →