How I've Actually Been Using AI

A lot of the AI talk online is about building brand new apps in a weekend.

That is part of it, but honestly, that is not the main way I have been using it.

The biggest value for me has been helping maintain and improve projects I already have. Old Laravel apps. Old browser extension code. Docs that should have been written a year ago. Small internal tools that are useful, but never feel important enough to stop everything else and build.

That is where AI has been the most useful for me.

Not as a magic button. More like a really fast assistant that can read code, trace things, explain what is going on, and help do work I would normally put off.

Converting old projects to newer tech

This has been one of the biggest wins.

Most founders have old projects sitting around that still make money, still get traffic, or still matter in some way, but the stack is outdated. Maybe it is an older Laravel app. Maybe it has jQuery everywhere. Maybe the frontend is a mess. Maybe the deploy process feels fragile.

That kind of work is easy to avoid because it feels like pure maintenance.

AI makes that work less painful.

I have been using it to go through older projects, understand how they are wired together, and help convert pieces to newer tech without starting over. That could mean moving a project to a newer frontend setup, cleaning up old patterns, or rewriting small parts in a way that is easier to maintain.

A lot of these projects do not need a full rebuild. They just need a path forward.

AI is good at helping find that path.

Framework upgrades are way less intimidating now

Another big one is upgrading Laravel versions.

If you have a few apps, this is the kind of thing that gets pushed down the list forever. It is not exciting work. But letting apps get too far behind is how maintenance turns into a bigger problem later.

AI helps a lot here because it can:

• look through the codebase for deprecated patterns
• compare current code to what newer Laravel versions expect
• suggest the likely breaking changes
• update config or syntax in a bunch of places faster than I would by hand
• help write tests or at least give me a checklist before I touch anything

It does not remove the need to review things. I still want to understand the changes.

But it turns a task that feels annoying and open-ended into something much more manageable.

Instead of thinking, "I do not want to deal with upgrading this app," it becomes, "Let me have AI map out the work, make the first pass, and I will clean up the rest."

That is a big difference.

Writing help docs from the actual code

One thing I think is really underrated is using AI to generate help docs by looking at your codebase.

For T.LY, that has been useful.

If AI can see the routes, UI, settings, and feature names, it can generate a pretty solid first pass of help center content. Not generic marketing copy. Actual help docs based on what the product really does.

That means I can have it write docs for things like:

• Custom domains
• QR codes
• Analytics
• Smart URLs
• Team features
• API-related setup
• Integration workflows

If you have an app and your help center is weak or outdated, this is low-hanging fruit.

Most of us already have the answers buried in the code. The problem is turning that into useful docs. AI is really good at helping with that part.

It can basically help generate a help center for your app if you give it the right context.

I still edit the final version because AI loves to sound too polished sometimes. But getting a solid draft from the actual product instead of starting from a blank page saves a lot of time.

Turning support into documentation

This connects to the docs point, but it is slightly different.

A lot of support issues are not really product bugs. They are gaps in explanation.

Someone asks how to connect a custom domain.
Someone is confused about link expiration.
Someone is not sure where analytics are.
Someone does not understand the difference between two plan features.

Once I see a few of those come in, AI can help turn that pattern into docs, FAQ answers, onboarding copy, or even better in-app text.

That is useful because it helps me close the loop faster.

Instead of answering the same thing over and over, I can use those questions to improve the product and the help center.

Understanding code I have not looked at in a while

This might be the most realistic use case of all.

Even when it is your own project, there are parts of the codebase you have not touched in a year or two. You vaguely remember how it works, but not enough to confidently change it right away.

AI is great for that.

I use it to ask things like:

• how does this flow work end to end?
• where does this setting get saved?
• what triggers this job?
• what parts of this feature are actually being used?
• what is safe to remove?

That saves a lot of time.

Instead of manually reading file after file trying to rebuild the mental model, I can get to the important part faster.

Writing one-off scripts I would normally avoid

Another good use case is small scripts.

Every app has boring cleanup work: data fixes, exports, imports, renaming things, checking records, backfilling data, cleaning old content, generating reports.

That stuff usually sits on the list because it is useful but hard to justify.

AI makes it easier to say yes to those tasks because it can help write the script quickly, especially when it has access to the actual models and schema.

I have found that helpful for all the little operational tasks that keep an app healthy but never feel urgent enough on their own.

What I am not doing with it

I am not blindly letting AI rewrite huge parts of a business-critical app and shipping it without looking.

I also do not think it replaces product sense.

It can help build. It can help maintain. It can help explain. It can help draft.

But you still need judgment.

You still need to know what matters.
You still need to decide what should be simple.
You still need to know when the AI output is making things worse instead of better.

That part is still on us.

My main takeaway

The most useful AI work for me has not been flashy.

It has been the boring, valuable stuff:

• upgrading old apps
• modernizing older projects
• writing help docs from code
• turning support questions into documentation
• understanding forgotten parts of a codebase
• writing small internal scripts faster

That is the kind of work that keeps products moving forward.

So while a lot of people are talking about using AI to start new things, I think one of the best uses is helping you finally clean up, upgrade, document, and improve the things you already have.

That alone is worth a lot.