Updating the revision description in one title block takes about a minute. Updating it on every sheet in a drawing set is the kind of job people hope an AI agent can take off their hands, and it's also where automation tends to go wrong. One drawing behaves the way you expect. Forty drawings collected across several years of a project usually don't.
AI agents for DWG drawing automation
AutoCAD AI tools can already handle useful work inside a drawing: answering questions about its contents, updating block attributes and pulling quantities. cadssistant, for example, publishes a recipe for building a bill of materials from drawing blocks. That's a good illustration of what an agent does well within one drawing. It reads the blocks, groups them and reports the result.
The difficulty is moving from one open drawing to a folder of them. Editing the drawing in front of you and processing many files unattended are separate capabilities, and some tools only do the first. Some AutoCAD connectors need AutoCAD to stay open while they work, which makes them interactive tools by design. Before you plan a batch, find out which kind of tool you have.
Why a folder is harder than one drawing
The drawing used in a demo is usually clean. A real set has history in it. Some sheets still use the title block from before the office standard changed, with different block names or attribute tags. A sheet number that's unique within each file can repeat across the set, so "A-201" alone doesn't tell the agent which drawing you mean. Some files contain two title blocks, perhaps an old one left in a corner of the layout or one on a second layout tab. In other drawings the attribute you want to fill in doesn't exist at all. And somewhere in the folder there is probably a superseded file that nobody deleted.
All of these are ordinary, and they decide whether a batch update saves time or leaves you with a cleanup job.
What a careful agent does before it writes anything
For each drawing, a careful agent identifies the title block it intends to change, reads the current revision value, and shows it beside the new value from your revision schedule. That report is your chance to catch problems while they're still cheap. A wrong mapping on a legacy block, or a sheet matched to the wrong row, is obvious when you see "current: Issued for Permit / proposed: Issued for Tender" next to a sheet that was never issued for permit.
Matching should use both the drawing and the object, so a repeated sheet number or block name can't send a value to the wrong file. When a drawing contains more than one candidate title block, the agent should apply the rule you gave it or ask you. It should never quietly edit the first one it finds.
Account for every drawing
After the run, every drawing in scope should fall into one of four groups: changed, already correct and left alone, skipped with a stated reason, or failed with an explanation. The four counts should add up to the number of drawings you started with. If they don't, some files went unaccounted for, and those are the ones you'll find later with an old revision on a plotted sheet.
Decide in advance how to handle the awkward cases. A legacy title block might get its own mapping from old attribute tags to new ones, or you might send those sheets back for manual work. A missing attribute usually means the block needs repairing. That's a separate change your revision update didn't authorize, so the right outcome is "skipped: attribute not present". The agent shouldn't invent a workaround.
Run it twice
Revision fields get updated again and again over a project, so the second run matters as much as the first. If you run the same instruction on drawings that are already correct, nothing should change. You shouldn't see duplicated text, an extra block or a second revision row. Then change one line in the revision schedule and run it again. Only the drawing on that line should change.
Interruptions are part of the same test. If the run stops halfway because AutoCAD closes, a file is locked or someone shuts the laptop, you need to know which drawings were saved and which weren't. Then a restarted run picks up the remaining files and doesn't apply the change twice to the ones already done.
Check the drawings, not the status message
"Batch complete" tells you the agent finished. It doesn't show you how the sheets look. Open the saved DWGs and look at a representative sample, including every legacy and exception case. When the fields you changed appear on issued sheets, plot a few and check them at sheet scale, where overlapping text and truncated values show up. Our title-block QA guide walks through checking a drawing set after edits like these.
If you want to try this with your own files, ArchiLabs Desktop works with native AutoCAD drawings and project files, and it can take the revision schedule spreadsheet in the same conversation as the drawings it applies to. Mapping rules for your legacy title block can be kept as project or team knowledge that colleagues can review and edit, so the next revision doesn't start from scratch. Start with a handful of sheets that includes the awkward ones, and ask for the current-and-proposed report before anything is written.
Audit project names, drawing numbers, revision fields, and sheet titles across a DWG set before issue.