Architecture, engineering, and construction firms have moved past asking whether AI matters. The live question is more practical: how does a firm turn scattered individual use into dependable work without weakening design judgment, client trust, or professional accountability?
AEC is in the gap between experimentation and operating capability. Chat, research, images, and drafting are common. Connected workflows over firm data—complete with owners, review gates, writeback, and measurement—remain the exception.
Executive summary
Five conclusions stand out from the combined evidence:
- AI use is common, but definitions hide the maturity gap. Surveys produce radically different adoption numbers depending on whether they count any tool use, regular job use, or firmwide implementation.
- Firms are using AI most where it is easiest, not necessarily where work hurts most. Chat, writing, research, and images lead; specifications, takeoffs, project controls, drawing intelligence, and connected knowledge remain less mature.
- Operational visibility is becoming a gateway use case. Firms want to combine pipeline, workload, project, financial, and delivery information into a current view they can question and act on.
- The real constraint is implementation. Data ownership, permissions, workflow design, professional review, administration, training, and adoption determine whether a promising demo becomes a firm capability.
- The frontier is moving from isolated tools to connected domains. The strongest systems retrieve evidence, draft or compare work, route it for review, record the decision, and update the system of record.
What this report is based on
This is a field report, not a claim to represent every AEC firm. It combines public industry research with direct observation of the questions, systems, and workflow problems firms are bringing into active AI work.
AEC AI Office Hours
Three practitioner sessions, attendance patterns, registration questions, live discussion, and follow-up requests from firms ranging from solo practices to large organizations.
Anonymized client audits
Workflow maps, systems inventories, stakeholder interviews, survey responses, pain-point analysis, solution portfolios, and implementation specifications across architecture, design-build, and construction firms.
Discovery conversations
Private discussions with firm leaders trying to understand what AI can do across design, delivery, finance, knowledge, growth, and management—not simply which tool to buy.
Industry and product evidence
McKinsey, AIA, buildingSMART, Autodesk, current product documentation, open-source workflow libraries, and case studies labeled by their evidence quality.
Two direct field signals are kept distinct. A practitioner pulse captured 16 responses about firm size, adoption stage, blockers, and risk concerns. Separately, 56 distinct people participated across three Clearworks AEC AI Office Hours sessions; 14 returned for more than one session. The survey describes only its 16 respondents, while the Office Hours count describes participation—not firmwide adoption.F1F4
Private firm evidence is anonymized and cited as Clearworks field research. It is used to describe recurring patterns, not to reveal a client or turn one firm's experience into an industry statistic. Vendor case metrics are identified as vendor-reported.
Adoption is broad. Capability is not.
There is no single honest “AI adoption rate” because the surveys are measuring different things. Autodesk's cross-industry 2026 survey says 98% of leaders use at least one AI tool.6 AIA's architecture-specific research paints a more layered picture: one-third of firms reported AI in day-to-day work in the 2024 Firm Survey, while a different AIA study found only 6% of professionals regularly used AI for their job, 8% of firms had implemented it, and another 20% were implementing it.34
Those numbers are not contradictory. They describe different thresholds:
| Threshold | What it can mean | What it does not prove |
|---|---|---|
| Someone used an AI tool | Chat, image generation, transcription, research, or an embedded product feature | That the firm supports or governs the use |
| AI appears in day-to-day work | Recurring individual or team use | That inputs, review, and accepted outputs are standardized |
| The firm implemented AI | Accounts, policies, workflows, or systems are being managed institutionally | That adoption is broad or business value is measured |
| A workflow is operational | Source, owner, review, destination, exception path, and measurement are defined | That it should become autonomous |
The gap is no longer awareness versus ignorance. It is personal use versus institutional capability.
That gap appeared clearly in a Q3 practitioner pulse. Of 16 respondents, eight were experimenting without a defined process, four had rolled AI out in one area, one reported a cross-team workflow, and three were not using AI or selected another status. Twelve of the 16 worked at firms with fewer than 15 people. These results are a directional snapshot of this practitioner group—not an industry prevalence estimate.F4
Firm leaders in audits and discovery conversations described the same underlying condition: paid business accounts used by only part of the team, committees formed to explore what comes next, staff experimenting with chat and renderings, and the recurring feeling that “there must be more we can do.”F2F3
Where firms are actually using AI
Current use clusters into three maturity bands. The bands matter because a workflow may be technically possible long before it is dependable enough for a firm to standardize.
Individual and bounded team work
- Chat, research, writing, and document summarization
- Meeting notes and candidate action items
- Proposal language and project-understanding drafts
- Image and rendering exploration from a controlled source
- Code and product research with professional verification
- Spreadsheet combination, analysis, and first-pass dashboards
Connected workflow systems
- Lead intake, pursuit research, and proposal assembly
- Project-report generation from PM and field records
- Open-loop tracking for RFIs, selections, submittals, and decisions
- Searchable project history and approved-detail reuse
- Model and drawing QA checks with human review
- Finance, workload, pipeline, and project-health visibility
Agent-led domains
- Multi-step design/documentation agents inside Revit
- Coordinated cost, schedule, procurement, and field-impact analysis
- Model-grounded accessibility and code review
- Decision-aware project records that update downstream plans
- Firmwide agent orchestration with approvals and audit evidence
- Robotics and physical-work automation on construction sites
AIA's 2025 specification study makes the mismatch visible. Among professionals with AI experience, chatbots were used by 79%, image generators by 50%, and grammar or text tools by 45%. Design and planning were at 13%, 3D modeling at 9%, and AI project management at 5%. Yet the same research identified product-list updates, estimating and takeoffs, complex specifications, and product research among the greatest inefficiencies.4
In other words, firms are often using AI where the interface is easiest while the expensive workflow friction remains deeper in the project and operating systems.
Production work is moving—carefully
The production-side story is more mature than “AI can make a picture,” but less mature than autonomous building design. Four patterns are becoming practical:
Start from something real
Teams use sketches, massing, Revit or Rhino geometry, model views, and existing renders to explore material, light, landscape, entourage, atmosphere, and presentation direction while controlling what may change.8
Automate repetitive setup
Emerging Revit assistants and specialist platforms can create views, sheets, tags, room data, schedules, and exports. These systems are best framed as bounded production assistance with traceable human review.9
Flag before issue
AI can help identify missing information, inconsistent tags, model-data gaps, accessibility conditions, drawing/spec coordination issues, and unresolved selections. Qualified professionals remain accountable for the finding and the final set.7
Make the project record queryable
Model properties, approved details, redlines, RFIs, submittals, decisions, and prior project records can become searchable, source-linked knowledge instead of memory held by a few senior people.
Safe verbs are draft, extract, compare, classify, flag, assemble, populate, query, and recommend. Architects and other qualified professionals author, verify, approve, and seal.
Why promising pilots stall
Across Office Hours and private audits, the same blockers appeared in different forms. The 16-response practitioner pulse made the pattern more concrete: seven respondents named uncertainty about where to start, four named limited time, budget, or R&D capacity, three named buy-in or change fatigue, and two named fear of choosing the wrong tool. Respondents could surface more than one constraint, so these counts describe mentions rather than mutually exclusive categories.F4
- The workflow is not defined.A tool is introduced before anyone agrees on the source, transformation, reviewer, accepted destination, and owner.
- The data is fragmented or untrusted.Project, finance, pursuit, meeting, and model information live in separate tools, spreadsheets, inboxes, and senior people's memory.
- The human is still the integration.People re-key consultant invoices, copy decisions into trackers, reconcile reports by hand, and chase status across email, chat, PM systems, drawings, and accounting.
- The experiment has no operating owner.No one owns access, connectors, templates, error patterns, support, training, measurement, or the decision to retire the pilot.
- Trust is treated as a feeling.Teams need explicit review gates, source citations, change logs, approved examples, and a place to record corrections.
- Client and project constraints arrive late.Healthcare, confidentiality, recording consent, IP, model-training, jurisdiction, and professional-duty questions must shape the workflow from the start.
- The firm adds another interface.A disconnected assistant can create more places to look. The useful layer reduces context switching and writes reviewed state back to the systems people rely on.
Risk and ownership compound those implementation problems. Fifteen of the 16 practitioner-pulse respondents named a liability-related concern, including professional liability, knowledge liability, or intellectual-property risk. Ten said institutional knowledge did not have a clear owner. That does not mean every firm shares those conditions; it does show why governance, knowledge ownership, and professional review have to be designed into the workflow rather than added after a pilot succeeds.F4
McKinsey describes the same shift at industry scale: firms create more value by redesigning end-to-end domains than by deploying isolated use cases.1 The unit of change is therefore not a prompt. It is a connected workflow.
Firm size changes the path, not the need
About three quarters of US architecture firms have fewer than ten people.3 Clearworks Office Hours also showed small firms as the strongest recurring known size segment, while large-firm attendance increased across each of the first three sessions.F1
Protect attention
Begin with the founder's recurring administrative and pursuit work, a secure supported assistant, and simple capture into an owned project record.
Reduce key-person risk
Connect handoffs, project status, proposals, finance, knowledge, and role-based training before fragmented habits harden into a larger operating problem.
Govern and scale
Standardize approved environments, domain ownership, data rights, evaluation, reusable workflows, and portfolio measurement while preserving specialized team needs.
The size gap in AIA's data is real: large firms report more AI use and implementation than small firms.34 But smaller firms may have a different advantage: fewer systems, shorter decision paths, and more direct access to the work. A small firm should not copy the setup of an enterprise lab.
The next step is helping the tools work together
The firm already has models, drawings, email, meetings, project software, finance tools, and a CRM. The goal is to help information move between them, make the next step visible, and keep the approved record where the team expects it.
This does not require one giant replacement platform. It requires enough connection and ownership that proposal work, project reporting, drawing review, dashboards, and knowledge search are not five separate experiments using five different versions of the truth.
Ask and capture
Chat or forms turn an inquiry, meeting, field observation, or question into structured candidate work.
Know what is happening now
Decisions, owners, due dates, changes, evidence, and problems stay connected to where they came from.
Reuse what works
The firm can find, compare, draft, and assemble work using its own rules and good examples.
Check it and save it
Qualified people decide. Approved work updates the PM, BIM, CRM, finance, or document system the team trusts.
What AEC leaders should do next
- Show people concrete examples. Let the team see possibilities across design, delivery, construction, finance, growth, and knowledge. Then choose what fits.
- Start small enough to own. One proposal, project report, drawing review, or billing process can matter without replacing how the whole firm works.
- Measure the work as it is now. Count time, rework, aging items, missing information, supervision, and trust before you change it.
- Name the person who checks it. Decide what AI may find, draft, compare, flag, or change—and who approves the result.
- Save the approved result. Every useful output should make the firm's project or operating knowledge easier to use next time.
- Plan for the people work. Accounts, privacy, permissions, training, office hours, corrections, and usage review belong in the plan.
- Let the second workflow reuse the first. Shared information, rules, review, and records are how isolated experiments begin to work together.
See the workflows. Then build the modern firm around them.
The Workflow Atlas shows what is possible. The Modern AEC Firm Playbook shows how the operating pieces fit together. A Clearworks Busywork Audit identifies the right sequence for your firm.
Sources and methodology
- McKinsey, “How AI is reshaping the future of the AEC industry,” July 15, 2026. Automation and value figures are projections; McKinsey analyzed more than 150 workflows across 25 AEC-related domains.
- McKinsey, “Delivering on construction productivity is no longer optional,” August 9, 2024.
- AIA, 2024 Firm Survey Report. More than 1,200 firms; firm-size and day-to-day AI figures.
- AIA, “Architect's Journey to Specification,” published 2025. This study uses a different instrument and population from the AIA Firm Survey.
- AIA Architecture Billings Index, May 2026 special question. 88% reported no staffing effect from AI in the prior year.
- Autodesk, 2026 State of Design & Make: AI Pulse. Vendor-sponsored survey across AECO, manufacturing, and media; not AEC-only.
- AIA AI Firm Toolkit, 2026. Current workflow and human-review guidance.
- Chaos, AI architectural visualization workflow with Enscape and Veras, June 2026. Product documentation; capabilities are not industry adoption evidence.
- Autodesk Revit 2027 Assistant technical preview. Preview status is material.
- buildingSMART, Industry Foundation Classes. Open, machine-interpretable data standards for cross-system workflows.
- Clearworks field research F1: anonymized AEC AI Office Hours attendance, registration, questions, and session records, July 20–August 17, 2026.
- Clearworks field research F2: anonymized architecture, design-build, and construction workflow audits, interviews, systems inventories, and solution portfolios, Q3 2026.
- Clearworks field research F3: anonymized AEC principal discovery conversations, Q3 2026.
- Clearworks field research F4: anonymized AEC practitioner pulse survey, Q3 2026 (n=16). Results describe the respondents and are not presented as a representative industry estimate.
Evidence labels: public industry survey; public product documentation; vendor-reported case; technical preview or proof of concept; anonymized Clearworks field research. Private evidence supports Clearworks' synthesis but is not independently inspectable by the reader.