I run a one-person company. Nine WordPress plugins, a couple of SaaS products, and no team to hand things off to. The only way that math works is leverage, and right now that leverage is AI tools. Here's the actual stack I use to ship — not the aspirational one.
This is the post I wish someone had handed me. Real tools, why each one earns its place, and one concrete way I use it.
Coding
Claude Code. This is the center of everything. It lives in my terminal, reads my whole repo, and edits files directly instead of making me copy-paste from a chat window. The difference is that it works inside the project, so it understands context I never have to re-explain. I use it for the boring 80% — wiring up an admin settings page in a plugin, refactoring a tangled function, writing the tests I'd otherwise skip — so I can spend my attention on the 20% that actually needs me.
AgentsCamp. A stock Claude Code install is good. A configured one is a different machine. AgentsCamp is the directory I built for exactly this: curated, format-validated agents, skills, and commands you install with one command. I got tired of pasting the same prompts in every week, so I packaged them.
In practice it's npx agentscamp add agents/code-reviewer before I open a PR, a commit-message command so I stop writing "fix stuff," and a web-research skill that pulls real sources instead of guessing. Each one turns a paragraph of instructions I'd retype constantly into a single move. Start with one agent and one command that match something you do daily — that's the whole trick.
The model itself. Worth saying plainly: the tooling is only as good as the model under it. I lean on the strongest reasoning model available for architecture decisions and gnarly bugs, and a faster one for the high-volume, low-stakes edits. Matching the model to the job keeps both speed and cost sane.
Shipping and infra
Vercel. My SaaS front ends deploy here. Push to a branch, get a preview URL, merge to ship. As a solo builder I have no DevOps department, so I want the deploy to be the boring part. Preview deploys also mean I can show a half-finished feature to a beta tester without touching production.
GitHub. Source of truth and the spine the rest hangs off. Every repo, every issue, every CI run. Claude Code talks to it through the gh CLI, so opening a PR or checking a failing run happens without me leaving the terminal.
Cursor and the CLI together. Most days I'm in the terminal, but when I want to see a diff land in context or scrub through a file visually, having an AI-aware editor open alongside Claude Code covers the gap. The point isn't loyalty to one tool — it's that the same project is open in both, and I reach for whichever fits the moment.
The theme here: infra you don't have to babysit. A solo team can't afford a stack that needs a stack to run it.
Writing and content
Code is half the job. The other half is making sure anyone can find the thing I built.
Claude Code, again. It writes first drafts of docs, changelogs, and release notes straight from the diff. I'm not handing it the keys to my voice — I edit every published word — but a rough draft from the actual changes beats a blank page every single time.
OptimizeCamp. This is the product I built because I needed it. Search isn't just Google anymore; people ask ChatGPT and Perplexity, and those engines decide what to cite. OptimizeCamp checks whether my content is actually readable and citable by AI engines — clear structure, answerable questions, the signals that make a page quotable. It's the practical side of generative engine optimization: instead of guessing whether an AI will surface my post, I get a read on it before I publish.
A plain writing setup. Markdown files in the repo, version-controlled like code. No heavy CMS, no lock-in. This very blog is markdown rendered by a custom Next.js pipeline. Writing where I already work means there's no context switch between shipping a feature and writing about it.
How it fits for one person
None of these tools are impressive in isolation. The leverage is in the seams.
A typical loop: Claude Code builds the feature, an AgentsCamp code-reviewer agent catches what I missed, Vercel ships a preview, I write the post in markdown right there, and OptimizeCamp tells me whether an AI engine will actually surface it. One person, but the workflow has the shape of a small company — a builder, a reviewer, an ops person, and a content team — except they're all tools, and they're all running in the same window.
The trap with AI tools is collecting them. I've installed plenty I never opened again. The stack that works is the small one you actually run every day, where each piece removes a specific bit of friction you felt last week. Add tools the way you'd hire — only when the pain is real and named.
That's the honest stack. It changes a little every month as the tools get better, and I build in public, so you'll see it shift. But the principle underneath stays put: as a solo builder, your job isn't to do everything yourself. It's to assemble a set of tools that does the everything with you — and then go make the thing only you can make.