All Things AI · Season 3 · What You Can Build · Episode 5
The tool you keep wishing existed can be built in an afternoon, described in plain English, and it belongs to you afterwards. This is the single largest shift available to an agent right now, and it does not require learning to code.
Everyone in this industry is renting something that almost works
Think about the software you actually pay for. Most of it does about seventy percent of what you need. The other thirty percent is either missing, or it is there but shaped wrong, or it is on a higher tier.
So you build a workaround. A spreadsheet beside the CRM. A naming convention that holds a system together. A Sunday evening habit that exists purely because the software will not do the thing.
The reason this is universal is that the software was built for a category of user rather than for you, and the economics of software have always meant that anything specific to how you personally work was never going to get built. That constraint is the thing that just disappeared.
What building actually means now
You describe what you want in plain English. The model writes the thing. You run it, tell it what was wrong, and it fixes it. That loop repeats until the tool does what you said.
This sounds like a simplification and it is not. The description is the skill. Knowing exactly what you want, in what order, with what edge cases, is the entire job, and that is a thing agents are already good at because it is the same skill as writing clear instructions for a person.
You are not learning to program
You are not going to read the code and you do not need to. What you are learning is how to specify a job precisely and how to tell whether the result is right. Those are both things you already do every week.
The tools are small on purpose
Nobody is building a CRM here. You are building the twenty-line thing that takes this export, pulls these four fields, formats them the way you actually need, and saves them there. Small tools that do one job exactly right are where all the value is, and they are what nobody could ever justify building for you.
It is yours permanently
Once it exists it is a file. No subscription, no seat, no vendor deciding to sunset it, no price increase. In three years it will still run. Very little else in an agent’s software stack can make that claim.
The kind of thing worth building first
Start with something small, repetitive, and annoying. Not something impressive.
- Anything where you take data out of one system, reshape it by hand, and put it into another. This is the single most common category and the most immediately satisfying to eliminate.
- Anything you do on a schedule that requires you to remember to do it. A tool that runs on its own removes both the task and the remembering, and the remembering was the expensive part.
- Any report you assemble by hand from the same sources every month.
- Any check you keep meaning to run and do not, because running it means opening four things.
How the first build actually goes
Describe the job as if you were handing it to a competent new assistant. What comes in, what should come out, what to do when something is missing, and where to put the result.
The model produces something. Run it on real material immediately, not on a made-up example, because the made-up example always works and the real material never does the first time. Something will be wrong. Tell it what was wrong in plain language.
Three or four rounds of that and it works. The first build takes an afternoon mostly because you are learning what to specify. The second one takes twenty minutes.
The rule that keeps this safe
Nothing you build gets to delete or overwrite anything on the first run.
The pattern is straightforward and it should be automatic: the tool writes its results to a new file rather than changing an existing one, and it shows you what it is about to do before it does it. This costs nothing and it removes the entire category of failure where a small tool eats something important.
Run it against a copy for the first week. This is not timidity. It is how you discover what it does at the edges, and the edges are where every problem lives.
Where this goes wrong
Three predictable failures, none of them technical.
Building too big for the first one. The instinct is to solve everything at once, and a large first project fails in a way that is hard to diagnose and easy to abandon. Build the twenty-minute thing. Then build the next twenty-minute thing. Six small tools that work beat one ambitious one that nearly does.
Describing the outcome instead of the process. Make my follow-up better is not a specification. Take this list, find everyone with no contact in ninety days, and write me the list with their last note attached is a specification. The gap between those two sentences is the whole skill.
Never running it on real material until the end. The made-up example always works. Real data is missing fields, has inconsistent formats, and contains the one row that breaks everything, and you want to meet that row on the first day rather than the thirtieth.
What it grows
Ownership, in a sense that is worth taking seriously.
An agent whose systems are rented has a business that can be repriced by other people. Every tool is a subscription, every subscription is a decision somebody else makes, and the accumulated stack is a fixed cost that only ever goes one direction.
An agent who can build has something different. The tools are assets, the cost curve is flat, and the specific advantages built into those tools are not available to anyone who simply bought the same software. That is a real competitive position, and it is one almost nobody in this industry currently occupies.
Why this site teaches it rather than sells it
There is nothing here to buy, and that is deliberate rather than modest.
A system you rent from someone is a system that can be taken from you. It can be repriced, discontinued, or changed in a way that breaks how you work, and none of those decisions are yours. The whole point of teaching this is that the thing you end up with is not dependent on anyone continuing to offer it.
That means the episodes are built so you can go and do it on your own accounts, with your own files, and keep it. Every one of them ends with you owning something rather than subscribing to something.
What it costs
Nothing per tool. Building runs against a model you are already paying for or one running locally for free, and the result is a file on your disk. The only real cost is the first afternoon, and it is paid once for the skill rather than once per tool.
Questions agents actually type
Can I build my own software without knowing how to code?
Yes. You describe the job in plain English and the model writes it, then you correct it in plain English until it works. The skill you are actually using is specifying a job precisely, which is the same skill as writing clear instructions for a person.
What should I build first with AI?
Something small and repetitive rather than something impressive. Anything where you move data out of one system, reshape it by hand, and put it somewhere else is the ideal first project, because it is bounded and the result is immediately obvious.
Is it safe to run a tool AI built for me?
It is if you make it write results to new files rather than changing existing ones, and run it against a copy for the first week. That single habit removes almost every way a small tool can cause real damage.
Why build a tool instead of buying software that does it?
Because software is built for a category of user rather than for you, so it does roughly seventy percent of what you need and the rest becomes a workaround. A built tool does the specific thing exactly, has no subscription, and cannot be repriced or discontinued by anyone else.
Season 1 is Claude Code for Real Estate Agents, filmed and on its own hub. All three seasons: AI for real estate agents.