Claude Code isn’t a feature to demo, it’s a tool to run day to day on paid client work. This guide curates what that actually looks like in practice: planning before code, debugging with real client-supplied evidence, working around the rough edges, and where it fits into billable work.
What is the actual workflow?
Plan first, then implement, then verify, then commit or open a PR, the same discipline as any careful engineer, with Claude doing the research and drafting at each step. The value isn’t skipping steps, it’s compressing the time each one takes without skipping the judgment.
Planning before code
Claude Code’s Plan Mode does real research before writing anything: it can fetch API documentation, work through edge cases, and produce a written plan to review and refine before implementation starts. See Claude Code Plan Mode: from idea to MVP in 7 minutes for what that looked like on a real utility script.
Giving Claude eyes on a bug
Claude Code can’t watch a video, but a client’s screen recording is often the clearest bug report you’ll get. Converting it into a sequence of still frames lets Claude step through the recording visually instead of you describing what happened in text. See using FFmpeg to let Claude Code see a client’s bug for the exact command.
Working around the limits
Claude Code has real rough edges worth knowing about before you hit them. If your code lives on a different machine than the one you work from, the desktop app can run Claude Code there over SSH: see Claude Desktop runs on Linux now, which also covers SSH sessions. Large codebase analysis can hit memory limits and crash the session; see when Claude Code crashes: using Claude.ai as a fallback for codebase analysis for a workaround that splits analysis and execution across two tools. And two Claude Code sessions can’t natively coordinate with each other; see two Claude Code sessions can’t see each other, give them a shared file for a simple way around that.
Setting up context per repo
Claude Code’s output quality depends heavily on what it knows about your codebase going in. See setting up CLAUDE.md for a codebase for what belongs in that file and what doesn’t.
Claude Code in an agency workflow
Using Claude Code on your own projects and using it on billable client work aren’t the same problem: client work adds review requirements, security boundaries, and trust considerations. See Claude Code for agencies and client work for where it actually pays off and the guardrails that keep it safe.
Ops built on Claude
Claude Code isn’t only for writing code. It runs personal and business operations too: an automated morning email digest posted to Slack, a reusable task-list workflow built from a single conversation, and awareness of ecosystem shifts like the WordPress.com Claude connector that signal where AI-platform integration is heading.
Keeping up with the model
The underlying model keeps changing, and what it’s newly capable of matters for how you use it. See Claude Opus 4.6: what actually matters for developers for what changed and why it mattered at the time.
Where it fits
84EM runs Claude Code on real client engagements, not as a novelty but as part of how the work actually gets built, reviewed, and shipped. See AI integration and development to talk about bringing that workflow to your project.
