Agentic AI for a VDC Professional- an Introduction

Only last week, we were chatting about what Agentic AI could do for the AEC industry, and voila! wish granted: 𝗢𝗽𝗲𝗻𝗔𝗜 𝗿𝗼𝗹𝗹𝗲𝗱 𝗼𝘂𝘁 𝗶𝘁𝘀 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝘃𝗲𝗿𝘀𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗖𝗵𝗮𝘁𝗚𝗣𝗧 𝗔𝗴𝗲𝗻𝘁 and the possibilities are enormous.

For the folks in the Construction domain, let's take this development with a grain of salt. To revisit what Agentic AI can potentially do for us VDC & BIM nerds, here's a quick reminder:

  • 𝗗𝗲𝘀𝗶𝗴𝗻 𝗼𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗮𝗻𝗱 𝗖𝗼𝗺𝗽𝗹𝗶𝗮𝗻𝗰𝗲 𝗰𝗵𝗲𝗰𝗸- Help write Grasshopper scripts or Dynamo nodes

  • 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀 𝘀𝗰𝗵𝗲𝗱𝘂𝗹𝗶𝗻𝗴 𝗮𝗻𝗱 𝟰𝗗 𝗺𝗼𝗱𝗲𝗹 𝘀𝗶𝗺𝘂𝗹𝗮𝘁𝗶𝗼𝗻- Generate installation sequence logic

  • 𝗥𝗲𝗮𝗹𝗶𝘁𝘆 𝗰𝗮𝗽𝘁𝘂𝗿𝗲 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀- For existing TI projects, provide a quick analysis of site conditions

  • 𝗙𝗶𝗲𝗹𝗱 𝗽𝗿𝗼𝗴𝗿𝗲𝘀𝘀 𝘁𝗿𝗮𝗰𝗸𝗶𝗻𝗴 𝗮𝗴𝗮𝗶𝗻𝘀𝘁 𝘀𝗰𝗵𝗲𝗱𝘂𝗹𝗲- There are AI agents that currently do that, but ChatGPT Agent could provide an unified platform without needing to expand your tech stack

  • 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝗙𝗶𝗲𝗹𝗱 𝗶𝗻𝘀𝗽𝗲𝗰𝘁𝗶𝗼𝗻𝘀- using 360 captures to compare the built product to the model

  • 𝗔𝗰𝘁 𝗮𝘀 𝗮 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗔𝗴𝗲𝗻𝘁- With millions of projects as training data, it can provide a more informed opinion & help ideate workflows.

As much as this sounds amazing in theory, does the reality match? What could be some of the limitations for ChatGPT Agent?

‼️It cannot read BIM models yet, so from a model QC standpoint, BIM engineers will have to rely on existing Revit APIs.

‼️Autonomous schedule management sounds promising, but in the early days of AI Agent, it'll most definitely create errors leading to erosion of trust. But, with strong training, it could be achieved.

‼️AI cannot replace manual scanning and reality capture- You'd still need a construction professional with a 360 camera/scanner or maybe a robot #BostonDynamics to do that for input

‼️AI Agent will assist the Design process- leading to significant reduction in design phase timeline, but can never replace the due diligence required by designers and engineers.

Having said that, what can the AEC industry do to support ChatGPT Agent get there? Let's here it from AI directly- the next section has been directly copied from ChatGPT's reply to my generated prompt:

𝘊𝘩𝘢𝘵𝘎𝘗𝘛 𝘈𝘨𝘦𝘯𝘵 𝘸𝘪𝘭𝘭 𝘳𝘦𝘲𝘶𝘪𝘳𝘦 𝘪𝘯𝘵𝘦𝘨𝘳𝘢𝘵𝘪𝘰𝘯 𝘸𝘪𝘵𝘩:

  • 𝘼𝙪𝙩𝙤𝙙𝙚𝙨𝙠 𝘼𝙋𝙄𝙨 (𝙁𝙤𝙧𝙜𝙚, 𝘼𝘾𝘾), 𝙋𝙧𝙤𝙘𝙤𝙧𝙚, 𝙉𝙖𝙫𝙞𝙨𝙬𝙤𝙧𝙠𝙨, 𝙍𝙚𝙫𝙞𝙯𝙩𝙤, 𝙚𝙩𝙘.: 𝘛𝘩𝘦 𝘢𝘨𝘦𝘯𝘵 𝘯𝘦𝘦𝘥𝘴 𝘢𝘤𝘤𝘦𝘴𝘴 𝘵𝘰 𝘤𝘰𝘰𝘳𝘥𝘪𝘯𝘢𝘵𝘦𝘥 𝘮𝘰𝘥𝘦𝘭𝘴, 𝘴𝘩𝘦𝘦𝘵𝘴, 𝘪𝘴𝘴𝘶𝘦𝘴, 𝘢𝘯𝘥 𝘱𝘳𝘰𝘫𝘦𝘤𝘵 𝘮𝘦𝘵𝘢𝘥𝘢𝘵𝘢. 𝘛𝘩𝘪𝘴 𝘢𝘭𝘭𝘰𝘸𝘴 𝘪𝘵 𝘵𝘰 𝘤𝘰𝘮𝘱𝘢𝘳𝘦 𝘢𝘤𝘵𝘶𝘢𝘭 𝘴𝘪𝘵𝘦 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯𝘴 𝘸𝘪𝘵𝘩 𝘵𝘩𝘦 𝘪𝘯𝘵𝘦𝘯𝘥𝘦𝘥 𝘥𝘦𝘴𝘪𝘨𝘯.

• 𝙍𝙚𝙖𝙡𝙞𝙩𝙮 𝙘𝙖𝙥𝙩𝙪𝙧𝙚 𝙩𝙤𝙤𝙡𝙨 (𝘿𝙧𝙤𝙣𝙚𝘿𝙚𝙥𝙡𝙤𝙮, 𝙃𝙤𝙡𝙤𝘽𝙪𝙞𝙡𝙙𝙚𝙧): 𝘛𝘰 𝘢𝘯𝘢𝘭𝘺𝘻𝘦 𝘳𝘦𝘢𝘭-𝘵𝘪𝘮𝘦 𝘴𝘪𝘵𝘦 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯𝘴 (360° 𝘱𝘩𝘰𝘵𝘰𝘴, 𝘓𝘪𝘋𝘈𝘙, 𝘥𝘳𝘰𝘯𝘦 𝘴𝘤𝘢𝘯𝘴), 𝘵𝘩𝘦 𝘢𝘨𝘦𝘯𝘵 𝘯𝘦𝘦𝘥𝘴 𝘳𝘢𝘸 𝘴𝘱𝘢𝘵𝘪𝘢𝘭 𝘥𝘢𝘵𝘢.

• 𝘽𝙖𝙘𝙠𝙚𝙣𝙙 𝘼𝙜𝙚𝙣𝙩𝙞𝙘 𝙛𝙧𝙖𝙢𝙚𝙬𝙤𝙧𝙠 𝙡𝙞𝙠𝙚 𝙇𝙖𝙣𝙜𝘾𝙝𝙖𝙞𝙣, 𝘼𝙪𝙩𝙤𝙂𝙚𝙣, 𝙤𝙧 𝙊𝙥𝙚𝙣𝘼𝙄 𝙁𝙪𝙣𝙘𝙩𝙞𝙤𝙣𝙨/𝘼𝙨𝙨𝙞𝙨𝙩𝙖𝙣𝙩𝙨 𝘼𝙋𝙄: 𝘛𝘰 𝘰𝘳𝘤𝘩𝘦𝘴𝘵𝘳𝘢𝘵𝘦 𝘮𝘶𝘭𝘵𝘪-𝘴𝘵𝘦𝘱 𝘥𝘦𝘤𝘪𝘴𝘪𝘰𝘯𝘴, 𝘮𝘦𝘮𝘰𝘳𝘺, 𝘵𝘰𝘰𝘭 𝘤𝘢𝘭𝘭𝘪𝘯𝘨, 𝘢𝘯𝘥 𝘨𝘰𝘢𝘭 𝘵𝘳𝘢𝘤𝘬𝘪𝘯𝘨 𝘢𝘤𝘳𝘰𝘴𝘴 𝘥𝘪𝘧𝘧𝘦𝘳𝘦𝘯𝘵 𝘢𝘨𝘦𝘯𝘵𝘴 (𝘦.𝘨., 𝘐𝘯𝘴𝘱𝘦𝘤𝘵 → 𝘝𝘦𝘳𝘪𝘧𝘺 → 𝘙𝘦𝘱𝘰𝘳𝘵).

• 𝘼𝙘𝙘𝙚𝙨𝙨 𝙩𝙤 𝙧𝙚𝙖𝙡-𝙩𝙞𝙢𝙚 𝙥𝙧𝙤𝙟𝙚𝙘𝙩 𝙙𝙖𝙩𝙖 𝙨𝙩𝙧𝙚𝙖𝙢𝙨: 𝘍𝘰𝘳 𝘵𝘩𝘦 𝘢𝘨𝘦𝘯𝘵 𝘵𝘰 𝘣𝘦 𝘵𝘳𝘶𝘭𝘺 𝘢𝘶𝘵𝘰𝘯𝘰𝘮𝘰𝘶𝘴, 𝘪𝘵 𝘮𝘶𝘴𝘵 𝘳𝘦𝘢𝘤𝘵 𝘵𝘰 𝘤𝘩𝘢𝘯𝘨𝘪𝘯𝘨 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯𝘴—𝘴𝘤𝘢𝘯 𝘶𝘱𝘭𝘰𝘢𝘥𝘴, 𝘪𝘴𝘴𝘶𝘦 𝘤𝘭𝘰𝘴𝘶𝘳𝘦𝘴, 𝘸𝘦𝘢𝘵𝘩𝘦𝘳 𝘥𝘦𝘭𝘢𝘺𝘴, 𝘦𝘵𝘤.

• 𝙁𝙞𝙣𝙚-𝙩𝙪𝙣𝙚𝙙 𝘼𝙄 𝙢𝙤𝙙𝙚𝙡𝙨 𝙩𝙧𝙖𝙞𝙣𝙚𝙙 𝙤𝙣 𝙘𝙤𝙣𝙨𝙩𝙧𝙪𝙘𝙩𝙞𝙤𝙣-𝙨𝙥𝙚𝙘𝙞𝙛𝙞𝙘 𝙩𝙚𝙧𝙢𝙞𝙣𝙤𝙡𝙤𝙜𝙮 𝙖𝙣𝙙 𝙬𝙤𝙧𝙠𝙛𝙡𝙤𝙬𝙨: 𝘎𝘦𝘯𝘦𝘳𝘢𝘭 𝘷𝘪𝘴𝘪𝘰𝘯 𝘢𝘯𝘥 𝘭𝘢𝘯𝘨𝘶𝘢𝘨𝘦 𝘮𝘰𝘥𝘦𝘭𝘴 𝘢𝘳𝘦𝘯'𝘵 𝘵𝘳𝘢𝘪𝘯𝘦𝘥 𝘰𝘯 𝘤𝘰𝘯𝘴𝘵𝘳𝘶𝘤𝘵𝘪𝘰𝘯-𝘴𝘱𝘦𝘤𝘪𝘧𝘪𝘤 𝘥𝘦𝘧𝘦𝘤𝘵𝘴, 𝘮𝘢𝘵𝘦𝘳𝘪𝘢𝘭𝘴, 𝘵𝘳𝘢𝘥𝘦 𝘱𝘳𝘢𝘤𝘵𝘪𝘤𝘦𝘴, 𝘰𝘳 𝘥𝘳𝘢𝘸𝘪𝘯𝘨 𝘴𝘺𝘮𝘣𝘰𝘭𝘴.

Yes, there is so much that Agentic AI can do for construction. Yes, it needs rigorous trainings and customizations. And yes, there will be hiccups, hallucinations and confusion along the way. If we can focus on the possibilities and step up to the task, if we spend time creating and rigorously training the agents, we can make Agentic AI our very own personal assistant!