Welcome to JPECOM: notes from building with AI every day
What this blog is for, what we write about, and how we try to keep every post practical and honest.
JPECOM Team1 min read
JPECOM is a small tech team. We build our own apps, games and e-commerce tooling, and we run AI agents as part of that work every day. This blog is where we write down what we learn along the way.
What you will find here
- Agent workflows. How we split work between coding agents, how we write tasks so they can be checked, and where things go wrong.
- Cost and setup. Practical notes on subscriptions, CLIs and local tooling for people who pay for AI out of a small budget.
- AI for small teams. What is worth automating in a small business, and what is better left to a person.
- Updates. Short, careful notes on changes in the AI landscape that matter to builders.
How we write
We try to follow three rules:
- Describe what we actually did, not what a tool promises.
- Say clearly when something is an opinion or an observation from our own use.
- No invented numbers. If we cannot measure it, we do not quote a figure.
If a post helped you, or if you think we got something wrong, tell us through the contact page. Corrections make the next post better.
What we build
Alongside the blog we are building Tower, a local-first desk for running agent CLIs side by side with plans, checks and evidence for each task. We also help small teams with web development and AI deployment.
Thanks for reading. New posts land here first.
Related posts
AI Update, October 2026: Notable Shifts for People Who Build Products
Trends we have been seeing: CLI agents are maturing, subscriptions come in more tiers, local models are worth a look, and verification tooling is moving to the center.
4 min read
AI for Small Businesses: Where to Start Without Burning Money
A three-step roadmap for small businesses: audit your workflows, pilot a single process, measure before and after, and only then scale up.
5 min read
Prompt → contract → verification: how we make agent work checkable
Write the done-criterion first, decide what a machine can check versus what needs a human, and keep evidence receipts so agent work can be trusted.
4 min read