For a solo or tiny team, AI is the closest thing to hiring a junior teammate for every role at once: engineer, marketer, support rep, researcher. Used well, it widens what one person can ship. Used lazily, it floods the internet with forgettable slop. The difference is judgement.
Where it actually helps
- Building: scaffolding, boilerplate, tests, debugging, and learning an unfamiliar API far faster.
- Marketing: first drafts, turning one post into ten, keyword and competitor research, ad variations to test.
- Support and ops: drafting replies, summarising feedback, triaging the inbox, turning docs into an FAQ.
- Thinking: a tireless sparring partner for pressure-testing positioning, pricing and plans.
Keep a human in the loop
AI is a force multiplier, not an autopilot. It is confidently wrong often enough that you have to stay the editor. As Simon Willison points out, code hallucinations are the least dangerous kind of mistake, because they fail loudly, while the subtle logic errors slip through unless you actually exercise and test the work. Ship nothing you have not checked, and don't let AI talk to customers unsupervised in the early days.
Don't ship slop
Willison defines slop as AI content mindlessly generated and pushed on people who didn't ask for it. The antidote is ownership: put your name on what you publish and stake your credibility on it. AI that helps you think is leverage. AI output you forward unread is the failure mode.
Don't automate the unscalable too early
The manual, unscalable work that wins your first customers, the personal onboarding calls and the hand-written emails, is exactly where you learn what to build. Automate it once you understand it, not before.
Use AI to clear the grunt work so you can spend more time on the two things it cannot do for you: talking to customers and deciding what to build. Leverage is the goal. Slop is the failure mode.