The Solo Builder's Guide to AI Shipping
March 1, 2026
I ship updates across 8 projects every week. I’m one person. No team. No employees. Three kids.
This isn’t a story about being productive. It’s a guide for solo builders who want to ship real products with AI coding assistants — not toy demos, not tutorial apps, actual software that people pay for.
Here’s everything I’ve learned running this setup for the past year.
The mindset shift
The biggest change isn’t tooling. It’s role.
You’re not a developer who also manages the product. You’re a product manager who also reviews code. That distinction changes how you spend your time.
A developer’s instinct is to open the editor and start writing. A product manager’s instinct is to decide what gets built, scope it, hand it off, and review the output. When your AI assistant is the one writing code, the second instinct is the one that ships.
I spend maybe 20% of my day writing or editing code directly. The other 80% is scoping tasks, reviewing diffs, making architecture decisions, and deciding what matters this week. That ratio felt uncomfortable at first. Now it feels like the only way to run multiple projects.
The daily rhythm
My day has three phases. They’re consistent enough that I can describe them without checking my calendar.
Morning: scope. 8:15 to 8:45. I open each project, check where it left off, and write task scopes. A scope is a tight description of what needs to happen — specific files, specific behaviors, specific constraints. I cover this in detail in how I scope tasks. By 8:45, I have 6–8 Claude Code sessions running in parallel.
Midday: review. 9:30 to 11:30. I flip through tabs and review diffs. Some tasks are done in 10 minutes. Some take 30. I read every diff line by line — that part is non-negotiable. I accept clean work, correct anything that’s off, and sometimes re-scope tasks that missed the mark. My daily workflow post walks through what this actually looks like on a real Tuesday.
Afternoon: ship. 1:00 to 3:00. Merge branches, run final tests, deploy what’s ready, prep tomorrow’s task list. Sometimes I write code directly in the afternoon — things that need my judgment more than my time. Architecture decisions, tricky UX flows, debugging something an agent got wrong twice.
This rhythm gives me 6–8 meaningful commits per day across multiple projects. Not lines of code. Commits. Features, fixes, pages, integrations.
The toolchain
I keep this deliberately simple.
- Terminal: iTerm2 with 8 tabs. Each tab is a project. Full-screen, no split panes. I flip between them. Details in how I run 8 sessions at once.
- Claude Code: One instance per tab. Terminal-native, spec-driven. No IDE dependency.
- Git: Feature branches for everything. One branch per task. Easy to review, easy to revert.
- Specs: Markdown files that define what gets built. The spec is the handoff document between me and the agent.
That’s it. No project management software. No Notion boards. No Jira. I keep a markdown file per project with a “next up” section and a “done this week” section. When you’re the only person on the team, overhead is the enemy.
What to delegate
Not everything should go to the AI. Here’s how I think about it.
Fully delegate:
- Implementation of features you’ve already designed
- Boilerplate and scaffolding
- Database migrations from a defined schema
- Test suites for existing code
- Copy changes and content updates
- Refactoring with clear before/after specs
- Bug fixes with reproducible steps
Partially delegate (you stay involved):
- New UI components (delegate the code, review the UX)
- API integrations (delegate the implementation, verify the error handling)
- Performance optimization (delegate the profiling, decide the tradeoffs)
Keep for yourself:
- Architecture decisions — which services talk to which, where the boundaries are
- UX decisions — what the user sees and when
- Business logic review — the AI doesn’t know your business
- Security-critical paths — auth, payments, data access
- First implementation of a new pattern — set the pattern, then delegate repetitions
The rule of thumb: if you can describe the desired output in concrete terms, delegate it. If the task requires judgment about tradeoffs you haven’t resolved yet, do it yourself.
The weekly output reality
I want to be honest about this because the internet is full of inflated claims.
In a good week, running this workflow, I ship:
- 3–5 meaningful features across my projects
- 5–10 bug fixes or improvements
- 1–2 new pages or content pieces
- Various maintenance tasks (dependency updates, copy tweaks, config changes)
In a bad week — a kid gets sick, a client has an emergency, the scopes are off and I’m re-doing work — maybe half that.
This is more than I could ship as a solo dev writing all the code myself. Significantly more. But it’s not magic. It’s not “I built a SaaS in a weekend.” Scouter took months. Triumfit took months. The speed improvement is real, but it’s a 3–4x multiplier on a real workload, not an infinite one.
Common traps for solo devs
I’ve hit all of these. Repeatedly.
Perfectionism. The AI’s code won’t match your style. It’ll use different variable names, different patterns, sometimes different approaches than you’d choose. If the code works, is readable, and passes tests — ship it. You can refactor later. You probably won’t need to.
Over-speccing. There’s a sweet spot for spec detail. Too vague and the output is wrong. Too detailed and you’ve spent more time writing the spec than writing the code. A good spec for a typical feature is 8–15 lines. If you’re writing more than that, the task might be too big. Break it down.
Not trusting the output. You still need to review everything. But review doesn’t mean rewrite. Read the diff. Check the logic. Run the tests. If it works, merge it. I spent the first month second-guessing perfectly fine code because it wasn’t written the way I would have written it. That’s ego, not quality control.
Trying to run too many sessions at once. Start with 3. Seriously. Your bottleneck isn’t how many agents you can launch — it’s how many diffs you can review thoughtfully. I worked up to 8 over several months. If your review quality drops, you’re running too many.
Skipping the spec. When you’re in a hurry, the temptation is to type a prompt instead of writing a scope. “Just add a settings page” instead of specifying what settings, where the data lives, what the layout looks like. Every time I skip the spec, I pay for it in review cycles. The spec isn’t overhead. It’s the product decision. Write it.
The compound effect
The reason this workflow matters for solo builders specifically is the compound effect.
When you’re one person across multiple projects, every project competes for your time. Before this workflow, I could actively develop maybe 2 projects per week. The others sat idle. That meant progress was slow across the portfolio and nothing had momentum.
Now all 8 projects move forward every week. Not equally — I prioritize based on revenue potential and urgency. But nothing sits idle for long. That changes the math on what a solo builder can sustain.
If you’re reading this and running one project, this workflow will make you faster at that one project. If you’re running several, it changes what’s possible. The solo developer shipping multiple projects post goes deeper on that specific challenge.
The setup takes a week to get comfortable with. A month to feel natural. After that, you won’t go back.