Your First Build, the Deadline Keeper

Ask any lawyer what wakes them at three in the morning, and sooner or later you reach dates.

Not the law. Not the facts. The calendar. Almost every mistake in a practice can be argued, fixed, or forgiven, but a missed deadline cannot. Let a statute of limitations slip, and it does not matter how strong the case was. The case is gone, the client is harmed, and no motion in the world brings the day back. It is the one error that is final.

So it may surprise you that your very first build is a small tool to keep those dates straight. And it may surprise you more that this tool, the one guarding the most important numbers in your practice, will not use the AI you just set up. Not one bit of it. That is on purpose, and it is the most important lesson in the book, so let me say it plainly before we build a thing.

The first rule: some things are too important for AI

AI is a wonderful guesser. That is, at bottom, what it is. It looks at everything it has ever seen and makes a very good guess at what comes next. Most of the time the guess is excellent. But a guess, however good, is the last thing on earth you want deciding whether your client's claim is still alive.

Anything that must be exactly right, every time, with no exceptions, does not belong with a guesser. Dates belong here. So do math, money, and counting. These belong in plain, dumb code that does the same thing every time, cannot have an off day, cannot misread the question, and cannot make something up to please you.

Your Deadline Keeper lives entirely on the safe side of that line. The AI helps you write it. Once written, it is just arithmetic, the same arithmetic every time, with no AI anywhere near the answer.

What you are going to build

Picture a simple page on your screen. A box where you type a starting date, the day of the injury, say, or the day the papers were served. A short list to pick from: two years, thirty days, ninety days, whatever rules you live by. You pick one, and the page returns the exact deadline, spelled out, no guessing. If it lands on a weekend, it also shows the next business day.

That is the whole tool. Not fancy. The kind of thing that quietly saves a career.

How you build it without writing code

Here is the part that sounds impossible and isn't. You are not going to write this. You are going to describe it, in plain English, to an AI, and the AI is going to write it for you.

This is level-one work, the public and harmless kind from Chapter 4. You are asking for a plain date calculator. No client anywhere in the request, no secret, nothing private. So you can use a strong AI in the cloud for this with a clear conscience. Building a generic tool is exactly what the cloud is good for. The tool you end up with is what stays home.

Open a capable AI assistant and tell it what you want, the way you would tell a sharp young associate:

Write me a single web page that calculates legal deadlines. It should have a box where I enter a start date, and a dropdown where I pick a rule like "2 years" or "30 days" or "90 days." When I click a button, show the exact deadline date, written out in full. If the deadline lands on a weekend, also show the next business day. Make it one file I can save and open in my browser, and make it work without any internet connection.

That is the whole job. Notice you did not write code. You wrote a description, in your own words, the way you would explain it to a person.

The AI hands you back a block of code. It will look like gibberish, and that is fine, because you no more need to read it than you need to understand your car's engine to drive to work. You just need to run it.

Running it, in about a minute

Copy what the AI gave you and paste it into a plain text file. Save it with a name ending in ".html", something like "deadlines.html". Find that file on your computer and double-click it. It opens in your web browser, and there is your tool: the date box, the dropdown, the button. Type a date, pick a rule, watch it answer.

You just built software. Sit with the "wait, that was it?" for a second, because that feeling is what this whole book is trying to give you. The wall was never as high as it looked.

Now do the one thing that matters: test it

You are not finished when it runs. You are finished when you trust it, and you earn that trust by testing it. Run it through cases where you already know the answer cold, and through the ones designed to trip it up:

• Three or four deadlines you have worked out by hand a hundred times. Do they match?
• A leap year. Does February behave?
• A deadline that lands on a Saturday or Sunday. Does it roll to the next business day?
• A deadline that lands on a court holiday, the Fourth of July, say. Does it handle it the way your jurisdiction requires?

If anything comes back wrong, you do not fix the code yourself. You go back to the AI, tell it exactly what was wrong, "the deadline that lands on a holiday should skip to the next business day, and right now it does not," and it hands you a new version. You save it, you test it again. You keep going until it is right every time.

This testing is not a chore. It is the whole craft. This tool is in charge of dates, the one thing that has to be perfect, and plain code is its right home precisely because you can test it until it is exactly right and then trust it to do the same thing tomorrow, and next year, without an off day. An AI guessing at the answer could never earn that trust, which is why we kept it out of the answer.

What you just learned

Look at what happened. You used AI for the thing it is wonderful at, building a tool from a plain-English description. And you kept AI entirely out of the thing it is dangerous at, being the part that has to be right. That split, AI to build it and plain code to be exact, you will use for the rest of your building life.

You also did the thing that ends the fear for good. Not read about building. Built. There is now a small, private tool on your computer that you made, that works with the internet off, that cannot make a mistake up, and that quietly guards the most unforgiving numbers in your practice.

In the next chapter the model does real work. You will hand it a long document, a record, a deposition, a contract, and ask it questions, and it will answer with the page right there so you can check every word. That is where your private AI starts to earn its keep.