Every Sunday, an email lands in my inbox from an agent I built called Sentinel. It’s a short brief on what happened that week across my industry, from the companies BoutiqueOS competes with to the platforms and tools it connects to: what they launched, what they changed, and the one or two things I should actually pay attention to.
I don’t have to remember to ask for it. That small difference turned out to matter more than anything clever the AI does.
How most people use AI
Most people use ChatGPT, Claude, or Grok like Google 2.0: type a question, get an answer, close the tab. That’s useful, but it puts all the work on you.
- You have to remember to ask. The busy weeks are the weeks you don’t check, and those are usually the weeks something mattered.
- You have to give it the background, in detail. What you sell, who you compete with, what you decided last month. The detail in your prompt is the biggest difference between a generic answer and a useful one, and most people type a single line.
- You start over. Next week, you’re back at a blank chat box asking the same question.
The chat apps can do more than that, and they keep adding features like scheduled tasks and memory. But for a question I ask every single week, I wanted something built around my business.
What I wanted instead
Sentinel isn’t smarter than a chatbot. It uses the same kinds of AI models. The difference is everything around the model.
- It runs on a schedule. Nobody has to remember. Sunday comes, and the brief shows up.
- It knows my business. It works from a context document that describes what I sell, who I’m watching, and what I care about. When something changes, I update that document once, and every future brief reflects it.
- It builds on last week. The brief focuses on what changed instead of handing me another snapshot.
- It follows a format I chose. Same structure every week: what moved, why it matters, what to do about it. I can read it in five minutes.
Why I run it myself
Sentinel runs on the same small server my other software already uses, and it costs very little to run. A cheaper AI model does the sorting, and a stronger one only handles the part that needs judgment.
That’s the part people don’t expect. A simple agent like this isn’t an enterprise project. It’s a scheduled job, a context document, and a well-written set of instructions. Because it’s mine, I decide what it watches, I can change it in an afternoon, and the context about my business stays current instead of starting from scratch in a chat window.
I’m not going to share every detail of how Sentinel works. But the pattern isn’t a secret. I call it the Sunday brief test:
- Pick a task you need done every week. For me, it’s a research project that builds on itself week after week.
- Write down the context an expert would need to do it well.
- Put it on a schedule and have the answer delivered where you already look.
If you find yourself asking AI the same question every week, the answer isn’t a better question. It’s making the answer show up on its own.
It isn’t just for retail
The same setup works for almost any business that needs to keep an eye on something outside its walls:
- A real estate agent could get a Monday brief on new listings, price drops, and what other agents are doing in their area.
- A service company could track competitors’ pricing pages, reviews, and job postings.
- A boutique could watch what brands and nearby shops are promoting before the season starts.
How I built it
- Started with one task: each week, find out what changed across my industry (the companies BoutiqueOS competes with and the platforms it connects to) and whether any of it matters for us.
- Wrote the context document: what BoutiqueOS does, who we sell to, which companies and tools to watch, and what I care about (and what to ignore).
- Defined the brief: what moved, why it matters, what to do. One screen, no more.
- Built the job with Claude: gather what’s new, use a cheaper model to sort out the noise, use a stronger one to write the brief, and email it Sunday.
- Tuned it after the first few weeks: tightened what counts as “worth my attention” and dropped sources that were just noise.