WorkBuddy Can Organize Quantity Takeoff Tables, But Don't Expect It to Draw CAD

It was past 8 PM. The AC in the project department office broke, and I was scrolling on my phone when I saw an article with a loud title saying the strongest AI tool of 2026 can already help designers draw CAD. Hand sketches, real photos, renderings—all can become CAD line drawings with one click. I glanced at it, and my first reaction wasn't awe, but wanting to ask: Can this thing help me produce the rebar delivery schedule?

The next morning, the client was pushing for materials. On my desk was a pile of stuff: structural drawing PDFs, rebar detail Excel sheets, site photos sent via WeChat by the construction supervisor, and scans of several change orders. Previously, for this kind of work, I'd either check line by line with a calculator or find someone on Taobao to make the tables. I've used WorkBuddy for over a month, seriously learned CAD for over a month, and started touching Excel AI tools around the same time. I wouldn't call myself proficient, but I have indeed used it for some dirty work a few times.

Drawings can become lines with one click, but the site still lacks a reliable summary table

Those AI CAD tools look great in demos. Line drawings come out, floor plans can be colored, landscape sections can be generated. I admit, for people doing proposals, presentations, and renderings, this saves time. But for someone like me, covered in dust on the construction site, what I see every day isn't whether the drawing looks good, but exactly how much rebar there is, if the specs are correct, and if it can be delivered.

When I threw those scanned drawings into WorkBuddy, I didn't hold out much hope. Last week I wrote a post saying its recognition of drawing parameters was always garbled. This time was the same. It could read most fields in the tables, but was dumb regarding annotations in the drawings. One line said Φ12 estimated quantity, and it treated "estimated" as part of the quantity. Another table note said "pending verification," and it included it in the summary anyway. Looking at the results, my first reaction was: Stop the fluff, can you separate notes from parameters first?

This doesn't mean WorkBuddy is completely unusable. It handles tables and documents much faster than my crappy computer. I reissued commands, telling it not to interpret images, but to only extract table regions. I required output by four fields: Spec, Length, Quantity, Unit. Anything in remarks, explanations, estimates, pending verifications, or site adjustments should not enter calculations, but go into a "Needs Review" column. This command was much more useful than the initial "Help me identify rebar parameters in the drawings."

In my tests, after running a dozen files, about 80% of fields matched. The remaining 20% were mainly handwritten notes, frame obstructions, and the same rebar being called different names in different tables. For example, the structural drawing says "12 threaded," the purchase order says "Φ12," and the material list says "HRB400E 12." WorkBuddy can group some together, but it won't fully make engineering judgments for me. It's like a very diligent new data clerk who doesn't understand drawing conventions.

WorkBuddy's true value isn't understanding drawings, but extracting the numbers

Later, I stopped forcing it to chew on drawings and took another route. For key tables in the drawings, I screenshot them first, or have the supervisor export them as Excel lists. WorkBuddy is responsible for merging details from multiple Excels, PDFs, and screenshots, doing field mapping, and generating a master table. After exporting the master table, I use Excel AI tools for pivots, VBA for formatting, and finally turn it into a delivery plan.

This step is crucial. WorkBuddy isn't design software, nor a CAD plugin. It's more like an office data transit hub. I used to think most people fail with tools because they never gave the tools a chance. That's not entirely right, but it was definitely a pitfall I stepped in. Initially, I always tried to solve everything with one command, resulting in garbage. Later, I learned: let it do one thing at a time. Table recognition is for recognition, merging for merging, summarizing for summarizing, validation for validation.

I did a rough calculation with this rebar plan. Manually merging these tables used to take me at least an afternoon, constantly switching windows, eyes blurring. With WorkBuddy, if the initial commands are clear, I can get a verifiable master table in about twenty minutes. The saved time isn't for rest, but for checking the rows marked "Needs Review." It extracts scattered numbers from files, but I still have to sign off on the final decision myself.

This is also my view on this wave of AI CAD tools. Those tools solve expression efficiency; WorkBuddy solves data organization efficiency. They are not the same thing. AI in CAD 2026 can organize repetitive lines into blocks, SolidWorks is preparing virtual assistants, and Microsoft Research is studying natural language model editing. These directions are interesting, but they are still far from material reconciliation on a construction site. What truly blocks me isn't drawing lines, but the inconsistency in standards between drawings, lists, changes, procurement, and acceptance.

AI can help designers draw CAD, but the construction site needs someone who can extract numbers from tables, documents, and screenshots even more.

I don't like hyping "one-click generation" too much. In engineering, the worst thing is looking pretty on the surface but collapsing underneath. WorkBuddy's current drawing recognition ability is poor; it easily mistakes notes for parameters—that's a real shortcoming. But for table summarization, field mapping, dirty data cleaning, and multi-file merging, it can indeed hold up. Who is it suitable for? People chased by reports every day: material clerks, budgeters, construction supervisors, project data clerks. Who is it NOT suitable for? People expecting one-click construction drawings, people hoping it makes judgments for structural engineers, and especially people who throw in drawings and wait for precise rebar quantities.

Looking forward, I hope WorkBuddy takes half a step further. It doesn't need to steal CAD's lunch; it should first understand object semantics in drawings. Which segment is an annotation, which is the frame, which row is a table, which word is a note, which value can enter quantity calculations. Then connect this rule set with Excel, CSV, and purchase lists. Only then will it not just merge tables for me, but understand a bit of engineering context. Not yet, but the direction is right. Purely stacking parameters is worse than tuning the model; same with tools—stacking flashy features is worse than smoothing out one workflow link.

So this post isn't praising WorkBuddy, nor bashing AI CAD. I'm just looking at things with site standards. If it helps me stay up one less night, it's good. If it generates a beautiful line drawing with one click but leaves me verifying numbers for half a day, it's still too far from the site. Using WorkBuddy for over a month, it hasn't turned me into a designer, but it did pull me halfway out of a mud pit of spreadsheets. Whether AI can draw pictures isn't the key; whether it can cleanly extract scattered numbers from drawings and tables is what matters most to us site people.