September 8, 2026
How I Automated My Deployment Pipeline with an AI Pair Programmer
I used to spend a chunk of every Friday babysitting deploys for client projects — running tests, checking logs, manually promoting builds. This is the setup I landed on after a few false starts, using an AI coding assistant to write and maintain the pipeline scripts themselves.
The short version: I stopped treating the AI assistant as a code-completion tool and started treating it like a junior engineer I could hand the whole pipeline to, with review. It wrote the initial GitHub Actions config, caught two silent failure modes I'd missed in the old bash script, and now helps me update the pipeline whenever a client's stack changes.
Two things went wrong along the way. First, it happily wrote a step that deployed straight to production with no approval gate — fine for a personal project, not fine for client work, so I added a manual approval step back in. Second, it kept "helpfully" pinning dependency versions that were already out of date by the time I merged, so I now always ask it to check current versions before finalizing a config.
Net result: deploys that used to take 20-30 minutes of my attention now take about three, and I trust them more than I trusted my old script, because there's a test suite gating every step.
automationai toolsclient work
August 10, 2026
A Field Guide to Choosing the Right AI Tool for Your Small Business
Almost every consulting call starts the same way: "which AI should we use?" The honest answer is usually "it depends on the task," so here's the framework I actually walk clients through, with the questions that matter more than any feature comparison chart.
Before we talk about specific products, I ask three questions: what's the task actually costing you right now (time, errors, or both), who needs to review the output before it goes out the door, and what happens if the tool is wrong 5% of the time. Those answers narrow the field faster than any "best AI tools of 2026" list.
From there, I group tools into three buckets: general assistants for drafting and research, narrow tools built for one job (scheduling, transcription, customer support triage), and custom workflows that stitch a few of those together with your own data. Most small businesses only need the first two — the third is where I usually get called in.
The biggest mistake I see is picking a tool because it's popular rather than because it matches the actual task and risk tolerance. A tool that's great for brainstorming marketing copy is a bad fit for anything touching customer financial data, and vice versa.
ai consultingsmall business
August 1, 2026
Teaching Yourself to Code in 2026: What Actually Works
I get some version of "how do I learn to code" almost every week, usually from someone who's already tried and stalled out. Here's the advice I give in office hours, without the productivity-guru gloss.
The pattern I see most often: people start a course, get through the basics, and then stall the moment they have to build something without step-by-step instructions. The fix isn't a better course — it's picking a small, real project on day one and treating the course material as a reference you dip into, not a track you follow start to finish.
AI coding assistants have changed this in a genuinely good way and a genuinely risky way. Good: you can get unstuck in seconds instead of giving up for the night. Risky: it's easy to let the assistant write code you don't understand, which means you learn nothing and can't debug it later. My rule for students: never accept a suggestion you can't explain back in your own words.
If you only take one thing from this: build something ugly and small that solves a real problem you personally have. It will teach you more in a weekend than another two weeks of tutorials.
learning to codeteaching