I've spent the last few days deep in the weeds of AI agents. Not playing with pr
By Nicholas Hodge · June 11, 2026 · Curated by George's Blog
I've spent the last few days deep in the weeds of AI agents.
Not playing with prompts, not generating images, not seeing who can build the coolest demo.
Trying to make AI actually useful inside a business.
And honestly, that's where things get interesting.
The challenge isn't getting AI to answer questions.
The challenge is getting it to consistently interact with systems, remember context, follow processes, and produce reliable outcomes.
That's a completely different problem.
I've had moments where I thought, "This is incredible."
And five minutes later, moments where I wondered how it managed to break something so simple.
AI gets a lot more useful when you stop thinking about prompts and start thinking about processes.
The biggest wins I've found haven't come from getting a model to answer a question better. They've come from identifying repetitive tasks, making the necessary information accessible, and building a workflow that lets the AI actually do something useful with it.
That's where most of my experimentation has been lately.
Not trying to replace people. Just trying to eliminate some of the repetitive work that consumes time every day.
Still plenty of things that don't work the way I expect.
But every iteration makes the next one a little better.