Every few years something arrives that is supposed to change construction completely. Some of it does change things, quietly, in the office. Very little of it changes what happens between a person and a wall.
What the software is actually good at
The honest list is shorter than the marketing and more useful than the skeptics admit.
Writing is the obvious one. Scopes of work, change order language, warranty explanations, the follow-up email you have been meaning to send for three days. A model drafts those in seconds, and you edit rather than start from a blank page. For anyone who does their invoicing at 9pm, that is real time back.
Reading is the underrated one. Manufacturer manuals, torque specs, fastener schedules, code summaries. Being able to ask a plain question and get the relevant paragraph, with the source attached, beats scrolling a PDF on a phone in a crawlspace.
Then there is the routine question traffic. Which battery fits this tool, what is included in this kit, do you have it in stock. Those questions have exact answers and no judgment involved, which is exactly the shape of problem software handles well.
What it cannot do
It cannot look at a floor joist and know that the discoloration is old and dry rather than active. It cannot feel the difference between a screw that is seated and one that is about to strip the hole. It cannot decide whether to tell a customer that the cheaper fix is the wrong fix.
Those are not gaps in training data. They are gaps in being there. A photograph of a subfloor contains a fraction of the information you get by stepping on it.
There is also the question of who is responsible. When a decision is wrong, someone signs for it. Software does not sign. That single fact keeps the important calls with a person no matter how good the suggestions get.
The trap with new tools
Most trade software fails for the same reason: it adds a step. If a tool requires you to open an app, log in, and enter data that you already wrote on a scrap of plywood, it will be abandoned by the end of the month. Nobody is dishonest about this. It just does not survive a busy Tuesday.
The tools that stick remove a step instead. They take a photo you already took and file it. They read the email you already received and put the date on the calendar. Judge anything new by that standard and you will waste far less money.
A reasonable position
Learn enough to use it for the boring half of the business. Estimating drafts from your own past numbers. Proposal language. Summarizing a forty page spec into the six lines you need. Answering the same customer question for the two hundredth time.
Keep the field judgment where it has always been. Check anything a model tells you about load, code, or safety against a real source before it becomes a decision. The failure mode of these systems is confident wrongness, and confident wrongness is expensive when it involves a beam.
Judging a claim
Vendors are going to keep promising more than the software does. Two questions filter most of it. What happens when the answer is wrong, and who notices. If the failure is a slightly awkward paragraph you edit, the risk is low and the time savings are real. If the failure is a load calculation that goes into a build, the tool needs a verification step in front of it and a person who signs.
The second question is whether it works with what you already do. A tool that requires a new habit from everyone on the crew has to be dramatically better than the clipboard it replaces, because the clipboard has a hundred percent adoption rate and never needs a password reset.
The part that does not change
Someone still has to carry the material in. Someone still has to hold the board while the fastener goes in. Someone still has to look a homeowner in the eye and explain why the price went up.
The trades have absorbed cordless power, laser layout, and a phone in every pocket without losing the essential shape of the work. This is another tool. Useful, worth learning, and not a substitute for knowing what you are doing.