I Build Agents Around Real Life

A lot of people still talk about AI like it is a better chat box.

The bigger shift for me happened when I started connecting agents to the systems I already live with.

The house. The gym. The business. My training. My games. The small decisions that repeat every week.

I care less about impressive demos now. I care about tools that remove friction from a real day.

The solar agent

At home, I connected an agent to the solar inverter API.

Its job is simple: watch the battery, check the coming weather in Cebu, and help decide how aggressive the house should be with solar usage.

If the next few days look rough, especially during storm season, the agent can be more conservative so we keep enough battery as backup for a power outage.

That is the kind of AI I like.

No futuristic speech. No big dashboard just for show.

A narrow system watching a real risk in the background.

Yamira

Yamira started as a household finance problem.

I wanted a budgeting app for me and Stej that fit how we actually live in Cebu. Peso spending, card imports, reimbursables, bills, shared household context, mobile-first use, and the kind of small money decisions that normal finance apps never quite understand.

So I built it from scratch with AI as the builder beside me.

It is not a generic expense tracker. It is our household operating system for money.

That matters because the problem was never, “Can I build an app?”

The problem was, “Can I build the app that fits our life closely enough that we actually use it?”

Radagon Coach

Then there is the ridiculous one, which is also useful in its own way.

I play World of Warcraft as a Protection Paladin. So I built Radagon Coach, an agent that pulls my Mythic+ runs, looks at run data, compares my play against top Protection Paladins, and turns that into coaching points.

Deaths. Defensive usage. Talent choices. Interrupts. Route patterns. What better players are doing that I am missing.

It is video game coaching, yes.

It is also a clean example of the bigger pattern: take a specific domain, connect the data, compare against a higher standard, then turn the gap into action.

That is coaching.

The dungeon just happens to have dragons in it.

Subtero Iron Track

For CrossFit Subtero, I program the Iron Track.

The work is not only writing the training. The admin matters too. The program has to end up inside WodUp cleanly, with the right dates, movements, tracks, notes, supersets, and publishing state.

WodUp did not give us a simple API path for this, so I used AI to understand the workflow and build an uploader around the way the system actually works.

Now the program can move from my source file into WodUp with verification instead of manual copy-paste.

That is not glamorous. It saves time, reduces mistakes, and makes the programming process less fragile.

Which is exactly the point.

Flame & Finish

For Flame & Finish, AI helped me build much more than content.

Inventory. Stock inquiry. Approvals. Owner dashboards. Accounting workflows. Agent API access. Staff-safe views. A system that knows cost and margin must stay owner-only.

That work matters because software used to feel external to the business. You paid someone, waited, explained the same thing many times, then hoped the tool matched the way the business actually moved.

Now I can turn operating judgment into software faster.

The judgment still has to be mine. AI does not know how our staff works, what our customers ask, which mistakes are expensive, or where trust can break.

But once I know the shape of the work, AI helps me build the system around it.

Small agents, real jobs

That is the common thread.

The solar agent protects backup power.

Yamira helps our household make better money decisions.

Radagon Coach turns game data into training feedback.

The WodUp uploader removes admin drag from gym programming.

The Flame & Finish tools make the business easier to operate.

I have also built an Apify actor for competitor ad intelligence, an agent command dashboard, local agents for routing and memory, and smaller tools that exist only because one repeated task annoyed me enough.

I use AI to turn a clear understanding of the work into a tool that does one useful job.

That is where the leverage is.

Know the work. Build the agent around it. Keep the human judgment in charge.