The Loadout

What's Actually Open Right Now

People love to talk about AI tools in the abstract. "What's the best model?" "Which IDE is the future?" "Should I use this or that?"

Here's a better question: what's actually open on your screen when you sit down to build something?

Because the answer, at least for me, is kind of boring. And I think that's the point.

The Daily Drivers

Google's Antigravity. That's the IDE. If you're reading the notes on this site, you already know the tool that built them. Before this, it was VSCode with the RooCode extension. Honestly? They're mostly the same. I switched because Antigravity shipped and I liked what I saw. I stayed because I'm old and human and switching costs are real.

Claude Opus. This is the main coding model. In my personal opinion, the reasoning and output quality from this model is currently the most reliable and highest quality for actual development work. When I need code that works — not code that looks like it works — this is where it happens.

Google Chrome + the Gemini extension. Gemini 3 Pro, specifically. This one is sneaky useful. Quick, easy access to a model right in the browser tab. Say I'm working on a new version of a page — I'll pull up production in one tab, dev in the other, and ask Gemini to do a comparison as a team of professional reviewers and subject matter experts. It's QA if you want to be real lazy about it. And sometimes lazy is fine.

The Snipping Tool + MS Paint (or whatever free image editor is already on your machine). This sounds stupid. It is not stupid.

Being able to quickly screen grab something, draw on it — circle the thing that's broken, arrow to the thing that should change — and then just Ctrl+V it directly into Antigravity? That's one of the most powerful workflows I have. It's instant visual context. No "the button on the left side of the header, no the other left, no the one below that" — just a screenshot with a red circle. Done.

But sometimes you need to go further than a circle. Sometimes you need to rearrange things. Say you want to show the AI what a layout should look like: snip the header, paste it into Paint, then snip a button from a different page, paste that in too, use the select tool to drag things where you want them.

It doesn't need to look nice. It just needs to be clear.

When you're happy with it, use the Snipping Tool again to grab your mockup from Paint, and Ctrl+V it straight into Antigravity. Snip, compose, snip, paste. Two free tools, zero subscriptions, and the AI knows exactly what you mean.

On Linux you've got Pinta or KolourPaint. On Mac, Preview does most of it. The point isn't which app — it's that getting the right context to an AI often doesn't require a subscription or a design tool. The stuff already on your machine is usually enough.

The Workflow

If it's a new project, the process looks like this:

1. Start with Gemini Chat. Not coding. Just thinking. I use the normal gemini.google.com interface — Gemini Thinking or Pro — and just work out the requirements. What is this thing? What does it need to do? What does it not need to do?
2. Create the documentation. BRD, requirements, scope — whatever the project needs. This lives in the chat first.
3. Open Antigravity. New git repo. Create the doc folder. Upload the markdowns from that first chat.
4. Start building.

That first chat — the BRD round — doesn't get thrown away. I go back to it. You can attach a code repo to the Gemini conversation and ask it to do a review. Ask it to point out scope creep. Ask it how you've been doing at following the original intent of the project.

That's the part most people skip. They plan, then they build, and the plan goes in a drawer. The plan should be a living thing you check your work against.

The Philosophy (If You Can Call It That)

Here's what I actually believe about tools:

Use what works until it doesn't.

That's it. That's the whole philosophy. I use Chrome because it's there. I use the Snipping Tool because it's there. I use Antigravity because it works for me right now. There's nothing romantic about it.

Unlike physical tools — where a better wrench is just objectively a better wrench — AI tools are personal. The "best" model is the one that produces the best results for how you think. The "best" IDE is the one you stop fighting with. The "best" workflow is whatever gets context into the AI's hands the fastest.

I started with VSCode and RooCode. They worked. Then something better came along and I switched. If something better than Antigravity shows up tomorrow, I'll switch again. No loyalty. No identity attachment. Tools are tools.

The Actual Point

The reason I'm writing this down isn't to flex a setup. There's nothing to flex. It's a code editor, a browser, a chat window, a screenshot tool, and a free image editor. A child could use this stack.

That's exactly the point.

The barrier isn't the tools. It never was. The barrier is knowing how to give the AI what it needs — context, constraints, clear intent. A $2,000 setup with bad prompts will produce worse results than Notepad and a well-structured conversation.

Everything on this site — the notes, the glossary, all of it — was built with what I just described. Not because it's the best stack. Because it's the stack I had, and I knew how to use it.

If you can't find a tool that does what you need? Build one. It's just a chat away.

Time is the only thing you don't get back. Spend some of it on the people who matter. Thanks for reading — now go build something.