A construction AI software lab
Exploring AI-native construction software and agent workflows for architecture, engineering, and construction (AEC) teams. From the first feasibility question to the building in operation.
Our Focus: Exploring, experimenting with, and building AI-native applications that understand real-world construction processes, workflows, and decisions—from initial feasibility through building operations.
Built around the whole process01 / The construction continuum
Follow the work from strategy through operations and renewal. At each phase, map the requirements, resources, responsibilities, and dependencies. Then investigate where AI could provide valid and early signal, help remove blockers, shorten lead times, and reduce avoidable costs.
Explore all 12 phases of the construction lifecycle. The construction ontology uses the same phases and connects their activities, responsibilities, requirements, and project records. Work can overlap and iterate.
Establish the problem, intended outcomes, and whether construction is the right response.
An agent workflow to explore
“Compare building, adapting, and leasing against our stated needs. Identify the assumptions and unresolved questions.”
Potential output: an options brief with source evidence, explicit assumptions, and questions for the sponsor.
Test whether the proposed outcome can be delivered at the site and within the project constraints.
An agent workflow to explore
“Compare these site options against our requirements. Show the unknowns that could change the decision.”
Potential output: a source-linked feasibility comparison and an investigation list for qualified review.
Establish who delivers the project, how it is funded, and who can make each decision.
An agent workflow to explore
“Review the appointments and responsibility matrix. Which decisions or package interfaces have no confirmed owner?”
Potential output: a responsibility-gap review and decision requests for the authorized participants.
Translate the brief into coordinated, buildable information and measurable performance requirements.
An agent workflow to explore
“This equipment selection changed. Find the affected requirements and drawings, and prepare questions for the design team.”
Potential output: an evidence-linked impact review with affected records and unresolved design questions.
Secure the project-specific permissions and commitments required for the relevant work.
An agent workflow to explore
“For this work package, identify the required permits and utility approvals. Show the conditions and missing authorization evidence.”
Potential output: a referenced conditions register with missing-evidence inquiries for the responsible team.
Translate defined work into an agreed commercial and delivery basis.
An agent workflow to explore
“Compare these trade proposals against the requested scope. Separate exclusions, allowances, and unanswered questions.”
Potential output: a bid comparison with source references and a reviewable scope-gap list.
Make reviewed information, resources, materials, and access available when the work requires them.
An agent workflow to explore
“Which upcoming activities depend on materials without a confirmed delivery date? Prepare the follow-up list.”
Potential output: a dependency-based exception list with source dates and proposed follow-up for review.
Execute work safely against current requirements while preserving evidence of what occurred.
An agent workflow to explore
“Turn these field notes into a daily report. Flag potential RFIs and show what needs clarification before anything is issued.”
Potential output: a draft report, linked evidence, and separately proposed follow-up actions.
Demonstrate that installed work and systems meet the requirements applicable to acceptance and use.
An agent workflow to explore
“Check the inspection and commissioning requirements for this system. Separate recorded results, missing evidence, and unresolved review decisions.”
Potential output: a requirement-by-requirement evidence review for the designated inspectors and reviewers.
Transfer usable assets and knowledge while closing the obligations applicable to the project.
An agent workflow to explore
“Check the handover requirements for this system. Show missing manuals, incomplete training records, and items awaiting acceptance.”
Potential output: a referenced closeout review that distinguishes receipt, acceptance, and remaining obligations.
Use the asset and its project knowledge to support reliable service and ongoing improvement.
An agent workflow to explore
“Find the operating and warranty records for this asset. Connect its maintenance history and identify missing or conflicting information.”
Potential output: an asset brief with source versions, documented terms, operating observations, and visible information gaps.
Decide how an asset changes, continues in use, or returns materials and lessons to the next lifecycle.
An agent workflow to explore
“Assemble the existing asset records for this renovation. Identify what needs survey or specialist review before adaptation or material recovery.”
Potential output: a source-linked existing-conditions brief and investigation list for the owner and specialist team.
Select a phase to explore. These are proposed workflows, not live agent executions. Cost, schedule, contracts, and decisions connect every phase.
02 / The construction ontology
Our starting point is the construction ontology: a structured map of the things, processes, and relationships that make a project work. We want to build that understanding across the lifecycle, then test how AI can use it to improve decisions and execution.
Every connection should retain its source and revision.
“What is stopping this equipment from being installed, and what would resolve it?”
Explore how agents could trace the requirements, approvals, supply, delivery, access, and predecessor work behind the answer. Identify missing evidence, investigate the cause of a blocker, and prepare options for the team to review.
A principle across the lab
Use identifiable, current evidence relevant to the affected work. Keep confirmed facts, reported information, and AI inferences distinguishable. Identify missing or conflicting information.
Identify a concern before a decision or commitment becomes difficult or costly to change, with enough time for the team to investigate and respond.
Explain what changed, the supporting evidence, what remains uncertain, the affected work, and when a response is needed to preserve the available options.
Across design, procurement, construction, and operations, we would test whether a supported finding reached the responsible people early enough for an effective response.
03 / Inside the lab
Start with the ontology and work through it, one connected part of construction at a time. Use AI to investigate how the work happens, where it gets stuck, and which changes could help. Look for valid and early signal while teams still have time to respond. Test proposed improvements against project evidence.
KNOWLEDGE / RELATIONSHIPS / CROSS-LIFECYCLE
Explore how AI can help identify construction concepts, connect related records, and surface gaps in our understanding. Keep proposed relationships linked to evidence and open to practitioner review.
The question: Can we build a useful shared model of how construction works, including the exceptions and dependencies that matter?
REQUIREMENTS / DECISIONS / COORDINATION
Explore how AI agents can identify emerging constraints from current project evidence while teams still have time to respond, challenge avoidable dependencies, and prepare coordinated alternatives. Bring project professionals a supported decision about what could change and which work could proceed.
The question: Can AI distinguish necessary constraints from avoidable dependencies and help teams find a faster, less costly way to achieve the required outcome?
DESIGN / PROCUREMENT / PRODUCTION / DELIVERY
Examine design decisions, reviews, purchasing, fabrication, delivery, and site readiness. Look for valid and early signal that a commitment is at risk, and evaluate changes to specification, sourcing, and sequence.
The question: Which elapsed time is necessary, which is avoidable, and would shortening it advance the overall project?
SCOPE / RESOURCES / EXECUTION / OPERATIONS
Investigate unnecessary scope, impending rework, and changing material prices and labor availability. Use valid and early signal to prepare alternatives before purchases, fabrication, or installation narrow the options.
The question: Can AI identify a supported opportunity to reduce total cost without moving the problem elsewhere or compromising required performance?
DESIGN / CONTRACTS / COORDINATION / FIELD / CLOSEOUT
Explore how AI could carry requirements, contractual obligations, decisions, and unresolved questions across project phases and teams. Detect conflicting direction, identify who needs to know about a change, and prepare communication tied to the affected work. Track whether the information was received, understood, and acted on.
The question: Can AI help the right people understand and respond to consequential information before a missed handoff creates delay or rework?
CONSTRUCTABILITY / FIELD EXPERIENCE / OPERATIONS / LEARNING
Explore how AI could capture project experience, establish where it applies, and use early simulations to test better ways to perform the work. Practitioners review the alternatives, and measured results improve the next decision.
The question: Can AI use project experience to anticipate problems, test better ways to perform the work, and improve subsequent decisions through measured results?
04 / AI in project coordination
We explore how AI agents can help project teams review drawings and specifications, track RFIs and submittals, identify missing information, and prepare follow-up for the responsible team members. Architects, engineers, and construction managers retain responsibility for technical reviews, approvals, and issued direction.
Retrieve the relevant requirements, records, relationships, and current versions within the user's access.
Compare evidence, assemble a draft or proposed action, and expose missing inputs and assumptions.
Carry out permitted actions. Obtain the responsible person's approval where the workflow requires it.
Inspect what actually changed. Preserve the source links, action history, and unresolved work.
Valid and early signal should lead to a supported proposal while the team still has time to respond. A generated answer, an executed action, and a human approval are different events. Our software should make each one visible.
05 / How we build
The Software Factory explores AI-native development methods and models for the applications behind our construction experiments. Our first experiment will use Factory as the harness, connecting valid signals to defined tasks, reviewed changes, and production feedback.
Buildings, infrastructure, and the people who deliver and operate them. Map how the work connects. Investigate the friction. Test where AI agents can help reduce delays, rework, and avoidable cost.
Explore the research agenda