How freelance developers actually use Claude in client work

Most posts about using LLMs in dev work are either marketing fluff or stories of someone shipping AI-generated bugs in production. The pattern that actually works for freelance developers in 2026 is narrower and more disciplined than either extreme. ## What Claude is good for in freelance work Drafting boilerplate (test scaffolding, OpenAPI specs, Terraform skeletons). Explaining unfamiliar codebases when onboarding to a new client. Reviewing your own diffs for missed edge cases. Generating documentation from existing code. Producing first-draft client emails and scope documents. ## What Claude is bad for Writing complete features unsupervised. Making architectural calls. Anything where a subtle bug costs more than the time saved. Tasks the client expects fully by you (read your contract). ## A typical workflow Read the brief, write a one-page scope yourself. Ask Claude to challenge the design at 10x and 100x scale. Plan the work as small commits. For each commit, write the test first, then ask Claude for a draft implementation. Read and rewrite in your own voice before committing. ## Be honest with clients Many clients accept LLM-assisted work as long as the code passes review and the deliverable is yours. Some explicitly forbid it. Read the contract; if unclear, ask in writing before the first commit. ## Pricing Productivity gain does not justify lowering your rate. The bug risk, warranty, and final responsibility stay yours.