Building Real Things with AI

This is how I use Claude, and it’s shaped by how I work. I’m more of a product developer, so I’m almost never doing the same task twice. For you, the repetitive parts of your job will most likely get automated inside your own platform, with Claude doing the work behind the scenes. All of this changes every fucking day, so we won’t lock into any one tool. We build it so that when something better comes out, we get better with it.

The model is a generic tool

Any AI model by itself is a generic tool. Out of the box, it isn’t going to tell you how to play game theory and rule the world.

What makes it powerful is running it in loops. It acts, gets feedback, checks the results, uses tools, and keeps iterating until the work is actually done. Those loops are what turn AI into SI.

Skills sit on top of that. They can be custom or prebuilt, and they encode the repetitive or specialized parts of your workflow so you aren’t starting from scratch every time.

Claude vs. Grok

Claude stays ahead. The leaderboards shift every month, but it keeps landing at the top for coding agents and real building work. You can check the current rankings here: https://artificialanalysis.ai/agents/coding-agents

Claude is the sharper model for building things right now, but it doesn’t really remember you. Grok Bot starts weaker, but it compounds into a partner that actually learns how you work.

Claude vs. Claude Code

Regular Claude is for conversation and one-off questions. Claude Code is for real work on a specific case or codebase. You point it at a folder and it can read, edit, and run things in that context.

The difference is big enough that you should treat them as two separate tools. If you’re working real cases or building something, you want Claude Code.

Reverse engineering anything

You can reverse-engineer almost any software by letting Claude take control of the screen. Open the app you want to understand and let Claude click through the workflows while it watches what happens. It maps the processes, figures out the logic, and learns how the software actually works in practice.

From there, it can generate a repo that recreates the same functionality as custom software. Once the processes are understood, the loops and skills take over and you own the implementation. This is one of the strongest ways to stop depending on someone else’s software.

Boxed software is losing its grip

Reverse engineering is part of why boxed software is losing its grip. If you can copy what a product does and own your own version, there’s less reason to stay locked inside someone else’s. More software is going to be custom, built around one specific company’s processes, while boxed tools stick around for commodity jobs like payroll.

You build it around the team you have, the connections you have, the exact workflow you have, and your employees’ actual skill sets. You stop forcing people to fit someone else’s product.

It’ll all be in one platform for that specific company: their marketing, sales, customer communication, scheduling, billing, documents, employee training, and reporting, everything they need for their whole business.

Developers need vision

A developer has to have industry experience, or they have to be creative and have a vision. That’s really the deal. Without that, they can only do what the person with the vision tells them to do. Now it’s faster and better for the person with the industry experience to build the tool.

Where I see it making money

I see the money in custom software for companies whose processes aren’t smooth. It lets them provide a better service and gain a bigger share of the market. They can also build custom tools that help them use the data that’s already out there.

Each deal also sets up the next one. If I do a personal injury deal, I can turn around and use the framework from the first one to make custom software for somebody else, built specifically to their liking.

I’m quick to understand an industry. If I can go into a company and spend a couple hours meeting with them every few weeks, I can understand exactly what they need and get it to them. Then I can integrate the other features they needed anyways.

Field service software

Falcon Directional is a company with crews and employees in the field and millions of dollars of equipment. You’ve got confused managers, crews with broken equipment, and everything fucked up. So you build a simple phone app where employees report their progress and every equipment failure.

You just deleted a manager. Now you have a smooth operating company with better billbacks and more revenue. Then you can add systems and trainings on top of it.

In the first month of integration, they were 4x more productive and saved money by tracking billbacks. Revenue went from $600k to $1.25 million without any other changes in the business. Long story short, they stopped using it because the president didn’t make the crews use the app after we left. It only took a total of 4 minutes a day from each crew. It was super fucking simple. Revenue is back down to $500k, and the company is barely hanging on. BoreTracker.com is the boxed version I have out in the market now.

Industrial Internet of Things

I value Industrial Internet of Things because it’s hard goods combined with coding that AI can do. And the market for it is much more niche, but there’s so much of it out there. Sensors in the field, industrial computers, and computing to the cloud all make or save companies money in some shape or form.

This used to be something only companies with millions of dollars could do. Now I can do it for about 5% of what it used to cost.

Industrial Internet of Things coupled with the Falcon approach is extremely valuable, because the time it takes now is ridiculous. The quantum monitoring gas prototype we built took about two months. I could literally do the whole thing in three days now, and I’m a leadership major who knows nothing about coding.

It took Jordan two weeks to recognize the values from one sensor. I can do that in 90 seconds now.

There’s a lot of opportunity here between Industrial Internet of Things and custom software solutions.

AI slop

“People treat AI like a vending machine instead of a collaborator.”

Most AI output right now is slop. Generating content is easy, so people generate too much of it. You still need someone who understands the domain well enough to know what actually matters and cut everything else.

Most people still use it for emails

I’ve been to a handful of weddings in the last six months. Most of the people I still keep up with use AI for emails and a bunch of bullshit, and they have no idea how fast you can build real things now. I know one person from college using it the way we’re using it, and Ty knows one, Alex Creede. Beyond that, we barely know anyone else leveraging it in a serious way.