All Things AI · Season 2 · The Models · Episode 10
Cursor gives a model your whole project at once instead of one document at a time. It was built for developers and the capability is not specific to code: it is the difference between a model that answers about a file and one that works across everything you have.
What it is
An application you work in that has a model built into it, with access to every file in whatever folder you have opened. You ask questions and give instructions, and it can read across all of it and change things.
It is made for programmers, which is why almost no agent has looked at it. That is a mistake, because the capability underneath is not about code. It is about a model that can hold a whole body of material at once and make consistent changes across all of it, and that shape of problem is common in any business.
Why whole-project access is different
The difference between this and attaching a document is larger than it sounds, and it changes what kind of question is answerable.
It finds things you did not know to ask about
Asking where does this appear across everything is a question you cannot ask a chat window, because you would have to know which document to attach. Here it searches everything and tells you, including in the files you had forgotten existed.
It changes many things consistently
Update this across every place it appears is a single instruction rather than an afternoon. The consistency is the point: doing it by hand is where things get missed, and the ones that get missed are the ones that cause problems later.
It shows you what it changed before it changes it
Every edit is presented for approval, with the before and after visible. This is the safety property that makes whole-project access reasonable rather than alarming, and it is worth knowing it is there.
What an agent would actually use it for
Anything where you have a body of related material and the work spans it.
- Templates and documents you reuse, kept in one folder, updated consistently when something changes.
- Checking a large set of your own material for a phrase, a claim, or a number that has since changed.
- Building and maintaining the small tools from the episode on building your own, because this is the natural place to do that work.
- Any collection of text files you maintain over time, where consistency across them matters more than any single one.
- Reorganising a body of material where the change has to be made in fifty places or nowhere.
The thing to understand about it being for developers
It will look unfamiliar, and the unfamiliarity is cosmetic rather than functional.
It presents files in a panel, it has a chat beside them, and it has features you will never touch. None of that requires understanding, and you can ignore the majority of the interface permanently.
The part you use is the chat panel and the approval of changes. That is genuinely it. The learning curve is shorter than most agent software, and it is steeper only in the first twenty minutes when nothing looks like what you expected.
Where it is weaker
It is not the tool for a single document. Opening a whole application to ask about one file is more effort than it is worth, and a chat window is better for that.
It works with text. Material sitting in a database, a hosted system, or as scanned images is not what this is for, and getting it into text first is a separate job.
And it sends your material to a hosted model by default. There are settings that change this, and for anything client-attached those settings matter, so it is worth checking them before you open a folder that contains sensitive files.
What to try in the first week
Two exercises, both chosen to show the capability rather than the interface.
Put twenty or thirty text files you already have into one folder, open it, and ask a question that requires having read all of them. Where do these contradict each other. Which of these mention a particular thing. What is in this set that is not in that one. The answer arriving at all is the demonstration.
Then ask it to make one change across all of them and watch the approval step. Reading a list of proposed edits, file by file, before any of them happen is the thing that makes this feel safe rather than reckless, and understanding that workflow is what makes the rest of it usable.
Where this goes wrong
Three failures.
Opening a folder full of client material without checking where the model runs. This is the important one. The default sends content to a hosted service, and that is fine for your own templates and not fine for a deal file.
Approving changes without reading them. The approval step is the entire safety mechanism and it becomes a reflex within a day. When it is changing forty files, read the list.
Using it for things a chat window does better. It is not a general assistant. Reach for it when the work spans a body of files and reach for something simpler otherwise.
Why a folder of text files is worth having at all
This tool only makes sense if you have a body of material, and most agents do not, so it is worth saying what that would even be.
The material an agent accumulates mostly lives in places it cannot be worked with: inside a CRM, inside email, inside a portal, inside somebody’s software. Each of those is fine on its own and collectively they mean you own very little you can actually operate on.
A folder of plain text files is the opposite. Your templates, your standard explanations, the answers you give repeatedly, your notes on how you handle particular situations, the briefs you have written for models. All of it in one place, in a format nothing can lock away, that any tool can read.
Building that folder is worth doing regardless of which software you end up using, and it is the thing that makes several capabilities in this season possible rather than theoretical. It is also the closest an agent gets to owning their own operating knowledge in a form that survives changing systems.
How it sits next to the others
It is not really a competitor to the chat services, because it is a different shape of tool that happens to contain one. The comparison that matters is between working in a chat window and working somewhere the model can see everything, and once you have a body of material the second wins comfortably.
It also pairs directly with the episode on building your own tools. That work happens in a place like this, and having a model that can see the whole thing is most of why non-programmers can now do it at all.
The honest summary is that it is the most unfamiliar tool in this season and the one with the highest ceiling for anyone who maintains a real body of material.
What it costs
A subscription in the usual range for the version with the better models, and a free tier that is enough to find out whether this fits how you work. The cost question is less interesting than the settings question about where your material goes.
Questions agents actually type
Can a real estate agent use Cursor?
Yes. It was built for developers but the capability is not about code: it gives a model access to every file in a folder at once, which suits any body of related material where consistency across all of it matters more than any single document.
What is the difference between Cursor and a chat window?
A chat window works on what you attach. Cursor works across everything in a folder, so you can ask questions you would not know which document to attach for, and make one change consistently across fifty files rather than by hand.
Is it safe to open client files in Cursor?
Only after checking the settings for where the model runs, because the default sends content to a hosted service. That is fine for your own templates and not appropriate for a deal file, and the setting is worth confirming before you open the folder.
Do I need to understand code to use it?
No. You use the chat panel and the approval step, and you can ignore most of the interface permanently. It looks unfamiliar for about twenty minutes and the learning curve after that is shorter than most software agents already use.
Season 1 is Claude Code for Real Estate Agents, filmed and on its own hub. All three seasons: AI for real estate agents.