Accepted record
The version of an AI-assisted output that a responsible person has reviewed and approved for the firm’s project or business record. The accepted record—not the model’s draft—is what downstream work should rely on.
AEC AI & workflow glossary
What do these AI words actually mean?
AI language gets abstract quickly. These definitions explain the terms through familiar work, tools, drawings, handoffs, decisions, and responsibilities inside an AEC firm.
If a term does not help you make a decision, it is not helping yet.
Each definition explains what the term means in the firm, what work it touches, and what a person still needs to own or check.
Agent / AI agent
Software that can pursue a goal through multiple steps, often using tools or connected systems. In an AEC workflow, an agent still needs defined permissions, review points, error handling, and an owner.
AI governance
The maintained rules and operating practices for approved uses, data boundaries, human review, ownership, training, and the evaluation of new AI use cases.
See a practical example →AI operating environment
The accounts, workspace settings, permissions, connectors, approved tools, documentation, review rules, and support practices that make AI usable inside a firm.
AI workspace administration
The ongoing management of an organization’s AI workspace: ownership, users, offboarding, SSO, privacy and retention settings, model-training controls, connectors, and access.
Automation
A repeatable rule or process that moves information or performs a task with limited manual handling. Useful automation has a known trigger, owner, exception path, and destination.
BIM
Building information modeling: the digital model and associated information used to coordinate design, documentation, analysis, and delivery. AI workflows can assist with BIM data, but the model and standards remain controlled professional records.
Blueprint
Clearworks’ short audit of one documented workflow. It makes the operating model, friction, ownership, and practical first fix visible.
See a practical example →Busywork Audit
A structured examination of how work actually moves through a firm, where information and ownership break down, which opportunities are worth pursuing, and what should happen first.
See a practical example →Connector
A controlled connection that lets one system read from or write to another. Connector permissions determine what data and actions become available to an AI tool or automation.
Controlled workflow
A repeatable sequence with defined inputs, permissions, review gates, ownership, exception handling, and an accepted destination. It is a more useful unit of AI implementation than an isolated prompt.
See a practical example →Data boundary
The explicit limit on what information a tool, person, connector, or workflow may access, retain, transform, or share.
Design intent
The architectural purpose and decisions the project team is working to preserve through visualization, documentation, coordination, and construction. AI can assist the workflow; the project team owns the intent.
Drawing intelligence
The use of computational or AI-assisted methods to query, compare, organize, or review drawings and models—for example, finding missing information, inconsistent tags, or revision drift.
See a practical example →Embedded Advisory
Clearworks’ ongoing support for AI administration, practical governance, training, adoption, workflow improvement, and the current roadmap.
See a practical example →Firm intelligence
A connected view of the information leaders need to run the firm, such as delivery, capacity, money, pipeline, follow-up, and operating exceptions.
See a practical example →Firm Operations System
The connected operating layer above a firm’s existing tools: trusted context, current state, reusable workflows, human review, system writeback, memory, and measurement.
See a practical example →Grounding
Giving an AI system trusted source material or structured context so its output is tied to the firm’s actual information rather than generated from general model knowledge alone.
Human-in-the-loop review
A required point where a responsible person evaluates an AI-assisted result before it becomes an accepted record or triggers a consequential action.
Model context protocol (MCP)
A technical standard that lets AI applications connect to tools and data sources through a common interface. For firms, the important questions are permissions, data exposure, supported actions, review, and ownership.
Model-grounded rendering
A visualization workflow that begins with a sketch, massing study, BIM model, or existing render so AI-assisted iterations remain anchored to real project geometry and design decisions.
Modern AEC firm
An architecture, engineering, or construction firm whose leadership, workflows, technology, knowledge, business model, positioning, and business development work as connected capabilities appropriate to its size.
See a practical example →Project record
The maintained source of project decisions, requirements, issues, approvals, and deliverables that the team relies on. AI-generated drafts become part of the record only after the appropriate review and acceptance.
Prompt library
A maintained collection of prompts, examples, settings, and review notes for repeatable firm work. A useful library records context and ownership, not just clever wording.
QA/QC
Quality assurance and quality control. In AI-assisted drawing review, software may flag possible omissions or inconsistencies while qualified professionals evaluate each finding and remain responsible for the issued work.
Retrieval-augmented generation (RAG)
A method that retrieves relevant source material and supplies it to a model when answering or drafting. In AEC, it can make standards, precedents, decisions, and project knowledge searchable with source links.
Review gate
A defined point where a named person must inspect, approve, reject, or correct an output before the workflow continues.
Source attribution
The record of where a fact, decision, lead, document, or model output came from. Attribution helps people verify information and understand which work or relationship produced a result.
Structured data
Information organized into consistent fields and types so people and systems can interpret it reliably—for example, a project issue with source, owner, status, due date, and accepted decision.
System of record
The authoritative system where an accepted piece of information is maintained. A workflow should specify where the reviewed result lives rather than leaving it inside an AI chat.
See a practical example →Tool sprawl
The spread of overlapping tools faster than a firm can establish ownership, settings, training, review, and support. The problem is usually operating coherence, not simply the number of licenses.
Workflow owner
The person accountable for how a workflow operates, who participates, what standards apply, how exceptions are handled, and whether the result stays useful.
Writeback
The deliberate step that places a reviewed result into the system of record—for example, turning an accepted meeting decision into a project task or issue update.
See a practical example →Make the whole firm work betterThe Modern AEC Firm PlaybookSee how leadership, workflow, technology, knowledge, money, positioning, and winning work fit together.What firms are doing nowAI in AEC — Q3 2026What people are using, what is still experimental, and what keeps good ideas from sticking.Find familiar work91 things AI could help withBrowse proposals, production, delivery, finance, operations, knowledge, and growth.