Two Shifts That Changed How I Use AI in My Business This Year
I had a good conversation with my friend Jason at All Great Things this week — the kind that leaves you with more questions than you walked in with, in the best way. A few things have been turning over in my head since, and I want to write them down before they evaporate.
Two big ideas. They go together.
The MCP shift
For about a year I've been building my own AI agents from scratch — my own platforms, my own dashboards, my own little setups with an LLM bolted on. It works, but it's a grind. Spin something up, get it humming, come back tomorrow, something broke. Spend half a day figuring out why a scheduled job stopped firing. Repeat.
There's a different way, and I've been slow to fully embrace it: stop building your own platform. Put the tools INTO the LLM.
That's what MCP — Model Context Protocol — actually lets you do. Instead of writing the integration yourself, wrangling auth, and praying it keeps working, you wire your tools up as MCP servers and let ChatGPT or Claude be the platform. They've already solved that part. You plug in and have a conversation.
For a small operator like me — and I'd guess most of the people reading this — that shift is everything. The hours I've sunk into building infrastructure I don't really need to own are hours I should have been spending on the work clients actually pay me for.
If you're running your own business and you've been trying to "build an AI tool" for something, ask yourself first: can I just put my tool into ChatGPT or Claude via MCP and have a conversation? Most of the time the answer is yes, and you're done in an afternoon instead of two weeks.
The guardrail problem
Here's the catch.
The easier it gets to spin up automations, the easier it gets to chase the wrong ones. I come up with five new business ideas before breakfast. Some of them are good. Most of them are me being bored. Without something pushing back, all five get built — half-finished, then archived, then forgotten.
Jason pointed me to Matt Pocock recently — a software engineer turned teacher who's been publishing some of the most useful AI content I've come across this year. Not vibe coding. Actual engineering: modular design, separation of concerns, debugging discipline. His blog is at aihero.dev and it's worth a follow.
Matt published a skill called Grill Me. It's tiny — just a few lines — but it pushes back hard. You hand it an idea — a new feature, a side project, a business pivot — and it interrogates you about it until you've reached real shared understanding. Who's it for? What does it cost in time? In money? Is it actually a good idea or are you just bored? Matt calls it the AI version of rubber ducking, and that's exactly what it is.
You install it in one line:
npx skills@latest add mattpocock/skillsI dropped it into my workflow this week and it has already talked me out of two side quests. Ten minutes of conversation with Grill Me beats seventeen hours of half-built code. That's the guardrail.
How they fit together
The MCP shift gives you leverage. Grill Me — or whatever your version of it is — gives you discipline. You need both, or the leverage just lets you build the wrong thing faster.
For my own business, what that looks like in practice:
- New automation idea? It goes through a Grill Me pass before I write any code. Half the ideas die there, which is the point.
- When I do build, I'm reaching for the MCP-server-into-the-LLM pattern first, not the build-my-own-platform pattern. If ChatGPT or Claude can be the platform, let them.
- Matt's other lesson, which I'm still working on: build features as modules. Each module is one feature, knows how it connects to the others, gets built in its own space. The seven-thousand-line routes file I'm slowly untangling in one of my own projects is my evidence for why this matters.
The takeaway
If you're a solo operator or running a small shop, the move this year is to be aggressive about the leverage AND aggressive about the guardrails. Either one without the other gets you in trouble — leverage without discipline burns money on dead ends, discipline without leverage means you stay stuck doing everything by hand.
Plug in. Then slow down enough to ask whether you should.
Thanks to Jason at All Great Things for the conversation that pushed all this. Good people make you sharper.