At time of writing, Claude Desktop is taking the construction industry by storm. Its ability to read, update, and create new files of any kind feels like magic. Many, including this author, are simultaneously impressed, amused, and terrified about what comes next. Claude Desktop looks like an automated python script kitty under the hood, but is so well put together that it feels like it could easily do any entry level job anywhere.
Remember when people were talking about the "AI Bubble" in 2023? In June 2026, that feels like ancient history. I'll bet Anthropic is going to make some money off of this buzz (that is not investment advice).
Speaking of python scripts, perhaps the most interesting observation I hear my peers making is that Claude Desktop is not only faster but MORE ACCURATE than humans. In the past, I had attempted to get AI do my auditing workflows directly. I would get some good results, but I could never rely on AI to do critical tasks. It was too eager to please, it would make things up, it couldn't count or do math, and it couldn't handle more then a few dozen PDFs.
Claude Desktop works differently. Instead of processing documents with AI, it will usually write a purpose built script to accurately extract text, tables, or images from PDFs. If it hallucinates while building the script, the script will error out, and Claude will fix it the problem before I can react. In the end, I have a workflow that would have taken me hours, days, or weeks to build, and the results will be almost perfect. If the results aren't perfect, they were still far more accurate than my old process of making a pot of coffee, printing a few hundred inspection reports or MTRs, and transcribing the contents of the files into an excel spreadsheet.
Let's say, hypothetically, that AI is becoming more than a slop generator. It now has the tools to count, do math, fix it's own errors, and process any sized task efficiently. The questions for a construction quality professional, assuming we don't go extinct, are these:
- When is it appropriate to use AI in Quality?
- What is the proper way to use AI in Quality?
If your answers are "Never" and "Not at all", then I know some Zoomers who are coming for you job.
I have attempted to tamp down my own job anxiety and answer these questions accurately and honestly. I believe that a Quality Professional that wants to do their due diligence, operate with integrity, but still keep up with the latest tools, should consider these principles that I am calling the CEDAR framework.
C - Conservation
No change here. Documentation is still an essential part of construction quality. We never threw away inspection reports after summarizing them before, and we aren't going to start now. We need to preserve every test and inspection report, shipper, MTR, etc. to be used in a turnover or just for our own records. This is doubly true if we use AI. If the client questions some AI output, we must have source documents to backup the AI. Not because we are worried the AI will get fired, but because we are ultimately responsible for what we have AI do for us. If you are using something like Claude Desktop, make sure to conserve the actual "skill" that contains the python script so that it can be reverse engineered.
E - Energy
Not everything is a nail. Us construction workers should know better. AI inference is a powerful new tool, but if you ask an AI model to count repair welds in a CSV weld log, it will most likely make something up. The AI appears to be eager to please. Instead of using inference, have the AI write a python script to loop through that file and tally repair welds properly. This is what Claude Desktop does in the background when you give it a task like this. Not only does it feel like magic, but it is more accurate and far more energy efficient. You don't have to be an environmentalist to understand that efficient tools are better tools. To avoid generating the same python script several times, across several projects, consider building an MCP server. This will improve auditability as well.
D - Diligence
Now that we are saving an audit trail, we need to actually audit the AI. Considering we are ultimately responsible for the AI output that we request, and then hand over to someone else, we are responsible for making sure it is accurate. The depth and breadth of auditing required will depend on how critical the task is. Integrity demands that we can reasonably back up any information that we provide to others.
A - Accuracy
As in the CSV weld log example in the Energy paragraph, running a script to count strings in a file is more accurate than using AI inference. Similarly, AI tools, especially if enabled with script writing, can be far more accurate than humans at performing clerical tasks, transcription, and summarization. If AI is more accurate, faster, and more energy efficient than a human, then AI should be used for that task. More accuracy means a competitive edge, but it also impacts code compliance and public safety.
R - Responsibility
AI must not be used to approve or reject any installed work or inspection results in the name of a human. AI does very well at raising red flags for compliance issues, and it is very good at finding human errors in reports, but we cannot foist our legal obligations onto AI. It goes without saying, but having an AI bot forge our signature is pencil whipping. You are still ultimately responsible.
P.S. Yes, I used AI to come up with the acronym and the generate the images for this article, but the ideas are mine (human).