AI in AEC Project Workflows

🤖 Architectural Drawing Reviews With Multi-modal AI Agents! 📐 If you've been following along, I recently posted some experiments exploring how vision-language models could be leveraged to extract information and knowledge from architectural drawings. Today I'd like to share some progress on Archie - our AI copilot for building codes and standards! We finished up a prototype to demonstrate how enhancing Archie with vision capabilities enables AI-assisted drawing reviews! I see a lot of promise in developing multi-modal, domain-adapted LLMs for the AEC industry and using them to power agentic AI systems with a deeper "understanding" of our domain's text, 2d data, 3d data, and artifacts. What's New? 🧠 Integrated fine-tuned vision-language model for object identification and information extraction from architectural drawings ✍️ Unlocked AI-assisted review workflows and automation What can you do with this? ✅ Compliance Checks At Scale: Automatically verify that architectural plans meet regulatory standards for specific elements. 🔍 Detailed Analysis: Assisted analysis workflows on entire drawing packages. Extract, classify, verify, annotate. 💡 Recommendations: Generate actionable insights and recommendations to address compliance issues found in the plans. 📝 Missing Data: Identify missing information required for compliance. ☠️ Inconsistent Data: Find inconsistencies in drawing standards and information. 📃 Example Use Case In the video below, you can see how Archie reviews an architectural plan based on a specific user request to check the bedrooms against compliance requirements (in this case the National Construction Code of Australia). Archie identifies and extracts the relevant elements in the floor plan and any relevant information. Then it retrieves applicable compliance knowledge. With all this information, Archie can perform compliance analysis against requirements including natural light and ventilation standards, electrical outlet placement, and window sill heights. The response includes a detailed summary and recommendations. 🚧 Challenges, And How You Can Help! There is a severe lack of AEC-specific datasets online, limiting research and progress. We need industry partners to come together and share the volume of data needed to make meaningful progress toward AI systems specialized for the industry, and resources for data cleaning, anonymization, labeling, and model training. We're happy to do the work, but we need your support to do it. So if you want to help - and transform how you work on the way - don't hesitate to reach out! I’m excited to hear your thoughts and feedback on the potential applications that these technologies can power! Drop your comments below! #AEC #Architecture #Engineering #Construction #Innovation #Technology #AI #MachineLearning #DecisionIntelligence #LLM #Mircosoft #OpenAI

If a clash changes and no one understands the impact…   does the project actually move forward?   Clashes don't happen in isolation, they affect everyone else on the project.   And when design changes occur, those downstream teams relying on that information need to see it, in a way that’s accessible to them.   ACC's API's give access to each iteration of clash tests, which run automatically every time there is a model update.   💻 Using the low code ACC Connect, I extracted clash information through the Autodesk APIs, along with all the other project information linked to the clashes, and sent that data to OpenAI for analysis.   AI then generated a simple dashboard showing what’s changed with summaries of each issue, related items like RFIs, due dates, and anything that needs follow up.   💡 This extra context means clashes can be viewed from new perspectives, like “which RFIs in ACC are linked to clashes that have been updated?”   It also means that instead of manually checking statuses, the system can: ✅ Push updates straight back to the linked record ✅ Notify the people affected ✅ Deliver insights where they’re most useful — in context   It’s like a conventional report but with added colour, context, and accessibility through AI.   This is one of many use cases where AI becomes the “last mile” of delivery, connecting the power of the clash engine and the APIs to the people who actually need the insights.   I’d love to take this further, making it interactive through natural language, and enabling users to look back across many revisions to understand how clashes are progressing over time.   If you're interested in exploring more, I've shared the dashboard and sample data, along with the APIs I used to get it, here: https://lnkd.in/gRbCGpMq   Tools like ACC Connect have so much potential to speed up coordination and close issues out faster, eliminating those pauses and manual steps that hold projects back.   Take a look at how it can simplify automation and working with APIs: https://lnkd.in/gJwqQ-at   #AI #AEC #AutodeskPlatformServices

I didn’t come from the construction industry. But I understood something fundamental early on: if you make data accessible to AI, it can reveal patterns that far exceed human capacity. The idea was simple—but transformative: Don’t bring AEC data to AI; bring AI to AEC data. That conviction led us, in late 2023, to begin embedding AI agents directly into the API layer of Revit. We proved that it was possible then. Now, after two years of intensive R&D, we’ve fully architected the paradigm. The infrastructure is defined. The vision is executable. The innovation is patent-pending. It won’t happen fast, but we can now imagine—and deliver—a future where conversations like these happen daily: 1. Project Architect: “Kymerra, run a full code compliance check against the 2021 IBC and local ADA. Flag all violations—egress paths, fire-rated penetrations, accessibility clearances. Generate a report with element IDs and proposed fixes. Create a 3D view for each issue.” 2. Structural Engineer: “Kymerra, extract the architectural model, build the analytical model, export to RAM for seismic zone 4 analysis. Import optimized steel sizes back into Revit. Highlight changes over 10% in a dedicated view.” 3. MEP Engineer: “Kymerra, run clash detection across all project models in Navisworks. In parallel, perform a preliminary energy analysis in IESVE. Generate a ranked clash report and EUI summary. Email both to the project lead.” 4. BIM Manager: “Kymerra, compare this week’s model to last week’s. Quantify deltas by discipline, update the MS Project timeline, and draft a summary email to the client with rendered exterior progress shots.” 5. Cost Estimator: “Kymerra, perform quantity takeoff for all concrete and steel. Cross-check with our cost database, apply 15% contingency, and output a phase-by-phase cost estimate in Excel.” We are building Kymerra - the agentic middleware layer for AEC. John Stump #aec #contech #bim #kymerra #terrascape

I built a free AI Prompt Library for construction teams. And no, I'm not going to ask you to comment "PROMPT" to get this. It covers the workflows that eat up the most time: - Estimating & scope gap detection - RFI and submittal drafts - Scheduling & lookahead planning - Project communication - Change order & T&M documentation - QA/QC & punch lists - Contract review & risk flags Each section has two versions: 𝗤𝘂𝗶𝗰𝗸 𝗽𝗿𝗼𝗺𝗽𝘁𝘀: for when you need something fast (10-20 min saved) 𝗗𝗲𝗲𝗽 𝗽𝗿𝗼𝗺𝗽𝘁𝘀: for high-stakes work like final bids, owner-facing COs, and contract disputes (30-60 min saved) Copy. Paste. Add your project details. Done. The prompts are built to ask clarifying questions instead of making stuff up. Because the last thing you need is AI confidently inventing scope that doesn't exist. I see people gatekeeping basic resources behind engagement tricks every day on this app. That's not how I operate. If it helps you do your job better, you should just have it. PDF is attached to this post for you to download/share/whatever you want. Which workflow would save you the most time this week? #ConstructionAI #AIforConstruction #ConTech

𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗔𝗜 AI doesn’t replace traditional architecture frameworks, it enhances them. Take, for example, the TOGAF Standard's #ADM. AI can act as a force multiplier for each phase. 🔸 𝗣𝗿𝗲𝗹𝗶𝗺𝗶𝗻𝗮𝗿𝘆 𝗣𝗵𝗮𝘀𝗲: Rapidly scan and synthesize architectural documentation to highlight recurring pain points. AI tools also support capability assessment. Skills inventories and role descriptions can be analyzed to identify gaps in the team’s abilities. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗔: Simulate business scenarios based on real enterprise data. AI can model the impact of implementing predictive maintenance, intelligent customer service, or algorithmic procurement. AI tools can analyze stakeholder communication to identify sentiment trends and key concerns. This allows architecture teams to tailor the vision to what stakeholders care about. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗕: Ingest workflow logs, screen interactions, and system traces to automatically map how business processes actually work, not how they are documented. These real-world models make it easier to identify inefficiencies, bottlenecks, and opportunities. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗖: Assist by automatically profiling data sources to assess their readiness for machine learning and analytics use cases. On the application side, AI models can recommend integration points for new capabilities. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗗: Simulate various deployment architectures and predict performance characteristics. This is especially useful in balancing on-premise and cloud strategies or designing hybrid environments. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗘: Use-case prioritization can be supported with scoring models that assess feasibility, ROI, risk, and stakeholder alignment. AI design assistants can generate architecture artifacts: draft diagrams and interaction flows. This dramatically reduces the time required to prepare solution documentation. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗙: Creating and continuously refining dependency graphs that reflect system interconnections, change risks, and stakeholder constraints. AI tools can also simulate different roadmap paths. E.g., how would a regulatory change impact the timeline? 🔸 𝗣𝗵𝗮𝘀𝗲 𝗚: Monitor project progress and detect misalignments with architecture specifications. This operates in near real-time, integrating with project management tools. Architecture compliance reviews become continuous and intelligent. 🔸 𝗣𝗵𝗮𝘀𝗲 𝗛: Monitor change signals (evolving regulations, new technologies, etc.) and surface emerging trends, risks, or opportunities. Feedback from users of AI-enabled systems can also be analyzed at scale. Applying AI to the ADM is about elevating the practice of Enterprise Architecture. The use of AI accelerates execution without losing structure. The methodology remains the same. The difference lies in how intelligently, quickly, and adaptively it can now be applied. ADM inset: © The Open Group #EnterpriseArchitecture #EA #TOGAF #OpenGroup #AI

For decades, vertical SaaS in construction has focused on one thing: Digitizing the Status Quo. We took paper schedules and put them on a screen. We took physical folders and turned them into PDFs. We called it "digital transformation," but really, we were just creating high-tech storage units. The workflow enhancement, the connection of the physical and digital world with complex automation over the last few years that pushed for insights beyond just data, fueled the real digital transformation in the AECO market. AI changes the game because it moves from systems of record to systems of agency. In vertical SaaS—specifically for the AECO industry—AI doesn’t just store your data; it commands it. Here is why AI wins the vertical war: Deep Domain Moats: Horizontal AI (like ChatGPT) is a mile wide and an inch deep. Vertical AI built on proprietary datasets understands the nuance of a "change order" vs. an "RFI" and the legal/financial weight each carries. AI can't replace the why and the certainty that domain brings to the equation. Vertical AI captures institutional knowledge and turns it into a repeatable engine and advanced decision making but in the AECO world. The winners won't just sell software "seats"; they will sell project outcomes. Those outcomes can only be built of the train tracks of deep domain expertise and understanding of complex systems of systems that drive safety and reduce risk and go beyond project management or estimates..... #ConTech #ConstructionAI #VerticalSaaS #DigitalTransformation #AgenticAI

I talked recently to an AEC customer who used this metaphor: “No one builds an entire project with one robotic arm. Yet that’s often how startups pitch us their AI.” One powerful feature can be impressive in isolation. But when it’s not embedded in a system or data infra that holds the rest, the backbone is too often missing just yet. Over the past two years, AEC customers and also we at Foundamental have seen a wave of AI-first startups that build standalone capabilities. Construction and AE customers get to choose among AI-first companies offering a single capability: generating 3D geometries, generating take-offs, summarizing project documents, extracting insights from plans, and so on. Useful ! However, Shubhankar and I have found ourselves returning to one core question when thinking about AI in the project-based world we love: What will HOLD these capabilities over time? In many cases, what we are envisioning more clearly is not AI as a standalone product, but rather AI as a feature – one that eventually requires a more persistent structure to live within. That much many founders and investors will agree with, on a “vision” level. ** But construction customers begin to ask themselves this: What will hold all my AI? ** The interface where the reps happen – whether a CAD environment, a construction ERP, or a collaborative spreadsheet – is not the AI. It’s the authoring tool, the system of record, the data infrastructure, the ERP or ERP migrator or process miners. (we could elaborate here why we think that, but our article and Spotify below does a far better job at that than we can do in a few characters here :) ) Cutting it short: We’re drawn to the companies building that structure that will compound so much that it has the best chance to hold all the AI for its AEC customers. Not because AI isn’t valuable – but because AI becomes truly powerful when embedded into something where the manipulation and iteration with further human input is most likely to happen. If you’re an AI-first company with a non-obvious insight into how you will compound into said SoR, ERP, migrator, data infra or process miner - we’d love to hear how you’re thinking about becoming that structure. #AI #AEC #ConstructionTech #AuthoringTools #DataInfrastructure #SystemsOfRecord #VC

For six months I've hosted small, curated A.I. peer groups with senior leaders from 30+ construction firms: Skanska, Lithko Contracting, Chandos Construction, DesCor Builders, Gray, Sundt Construction, and many others. The vendor conversation has matured. Across most cohorts this spring, the same shift showed up where builders are no longer picking one A.I. vendor and committing for the next decade. They're building portfolios of A.I. tooling with a primary LLM chosen for ecosystem fit and governance, paired with selective use of other models where a specific team's work demands a different capability. Construction makes this harder than it sounds. Most firms run on Microsoft 365, which means an assistant that inherits the right permissions and sits next to project files. But A.I. is increasingly personal and certain functions have preferences for different models or tooling. The firms out in front are governing at the workflow level rather than the tool level. They set the principle, then let teams choose the stack. For example, a technology director at a regional general contractor walked his peer group through standardizing on Microsoft Copilot organization-wide in February. One month later, in the same room, he had an update. Marketing was now in Anthropic Claude under a walled environment because content work needed it. His tech systems manager was testing whether Claude could layer on SharePoint and M365 data. The standardization held, but the portfolio widened. A VP at a different general contractor described the same principle at the individual workflow level. On a market entry analysis, he ran the deep research in Anthropic Claude, handed the output to ChatGPT and asked it to find any factual or unfactual information, then sent the cross-checks back. The governance wasn't which tool. It was that high-stakes outputs get verified across at least two of them. Different altitudes. Same idea. Govern the workflow, let the tools follow. What does your firm's two-model policy actually say in writing? #construction #buildingconstruction #artificialintelligence #ai #constructiontechnology #constructiontech #technologyexcellence

Autodesk is redesigning their entire ecosystem for interconnected work... Forma is set to become the cloud, data, and AI layer for Revit. For example, you can literally move Forma data into Revit without manual file exchanges or multistep workflows, then just include contextual data like: - terrain - parcels - surrounding buildings All setup for you to easily run analyses directly inside Revit. Forma is the interoperability engine of Revit and eventually everything else under the Autodesk umbrella. Revit stays as the trusted environment that uses Forma as the cloud layer for site context, early design, analysis, collaboration, and data management. Autodesk Assistant is also now in Revit, where it can guide you and help do tasks across Autodesk products. And we're seeing that Autodesk is building the future around Forma/Revit first since Revit is most suitable to hold the production BIM model, documentation workflow, firm standards, and user base. Instead of allowing firms to abandon Revit for a new cloud platform, Autodesk is wrapping Forma around the tool Architect's already start on. Autodesk Assistant will act as the AI orchestrator that can eventually act across all Autodesk products. Revit becomes one connected client in a larger cloud intelligence system. Then a power user's skillset shifts from just drafting to understanding data flow, context, standards, automation, and interoperability. Any technician who embraces, adopts, works with and understands how AI, Cloud and modeling software are converging in AEC is going to become a powerhouse in high demand over the next decade. #AEC #BIM #Revit #Autodesk #AutodeskForma #Forma #AIAssistant #DigitalTwin #ConstructionTech #AECTech #FutureOfAEC #BIMTechnology #DesignTechnology #Architecture #Engineering