I've spent the last two years building products for clients ranging from solo founders to established businesses. Here's what actually matters.
Scoping is everything. The projects that went sideways weren't technically hard — they were badly scoped. The best thing you can do before writing a line of code is produce a one-page spec that both you and the client sign off on. It forces clarity and gives you something to point to when scope creep starts.
Technical stack choices should optimise for handoff, not your preferences. Your client will likely need to maintain or extend the project after you're done. Default to the most boring, well-documented option that solves the problem. I've learned this the hard way by shipping something elegant that no one else could touch.
The clients who value quality will find you through referrals, not marketplaces. Marketplace clients optimise on price; referral clients optimise on trust. Build for one good referral client early and treat that relationship like gold.
The loneliest moment in freelancing is mid-project, when the excitement has worn off and the end isn't yet in sight. Having a small network of other independent developers to talk to makes an enormous difference — both practically (code reviews, second opinions) and emotionally.
Finally: raise your rates. Consistently. The market for good developers is tighter than you think, and underpricing signals risk to serious clients rather than value.