Three months ago I started using AI for every part of my freelance workflow. Some parts I expected. Some I didn't.
What got faster (a lot faster)
- Writing boilerplate code — yes, the obvious one
- Writing tests — pytest fixtures, factory_boy models, the boring stuff
- Translating English to code — "give me a function that does X" works most of the time
- Writing documentation — README, docstrings, OpenAPI specs
What got SLOWER
- Choosing between libraries — AI gives you 3 options, each with 4 tradeoffs, and now you have to evaluate 12 things instead of 3
- Architectural decisions — the LLM confidently recommends a stack that won't work for your specific constraints
- Debugging subtle race conditions — the LLM sees the traceback and suggests 5 fixes, none of which are right
- Pricing conversations — AI can draft the email but the judgment call is still mine
What got STRANGELY different
The hardest part of freelancing used to be writing code. The hardest part is now the human stuff: managing scope, setting expectations, navigating ambiguity, deciding what NOT to build.
AI is great at the work. It's terrible at the meta-work.
What I'd tell a freelancer starting today
Use AI for the things you're already good at. Use it to go faster on the things that don't teach you anything new. Don't use it for the things that make you a freelancer in the first place: judgment, taste, scope management, client communication.
The freelancers who lose their jobs to AI are the ones whose entire value was "I can write code." The freelancers who thrive are the ones whose value was "I can decide what to build and how to build it right."
I keep my freelance toolkit (118 prompts for the meta-work) here: https://gobvan.gumroad.com/l/wypos ($19). Free sampler: https://gobvan.gumroad.com/l/skzza.
Top comments (0)