Claude Code dropped the cost of custom software low enough that businesses that previously couldn't justify a developer can now afford one. The next problem is figuring out who to hire.
This is the practical version. What to look for, what to avoid, and the specific questions to ask before you wire someone $5,000.
What "Claude Code expert" actually means
Claude Code is Anthropic's command-line tool that lets Claude read, write, and execute code on your machine. A Claude Code expert is someone who uses it as a primary engineering surface, not someone who has it installed and uses Claude.ai for chat.
The distinction matters because Claude Code rewards a specific kind of operator. Someone who can write a clear spec, structure a project so Claude can navigate it, set up the right memory and skills, and review what the model does instead of pasting and hoping. The skill isn't "I know how to prompt." The skill is "I run engineering work through Claude as the engine."
The five things to look for
1. They've shipped real production systems, not just demos. A demo works in good weather. A production system handles edge cases, fails gracefully, has logging, and runs at 3am on a Tuesday when nobody is watching. Ask for production deployments they've shipped, not screenshots of chats.
2. They have a methodology, not just a toolset. Anyone can list "Claude Code, Trigger.dev, Supabase" on a profile. The senior version has a methodology underneath: how they audit before building, how they scope to prevent creep, how they handle multi-stage rollouts, what their definition of done is. If the answer to "how do you work" is a list of tools, they're not who you're looking for.
3. They show their work in writing. Senior operators write. Blog posts, case studies, public reasoning. Not because they're trying to be influencers but because writing forces clear thinking and serves as proof of expertise. If the only public footprint is a one-page profile, you have no way to evaluate how they think.
4. They're picky about the tool, not the tool's evangelist. A real Claude Code expert will tell you when Claude Code isn't the right tool. They'll suggest a no-code alternative when the problem is small. They'll recommend OpenAI for embeddings or Gemini for visual review when those are the right calls. If everything is a Claude Code nail, the hammer is the brand, not the work.
5. They have a track record outside Claude Code. Claude Code is two years old. Anyone with serious depth has prior engineering or operations background. Ask what they were doing before Claude Code existed. The answer should reveal a real operator, not a recent course graduate.
The five red flags
1. Vague portfolio language. "Built AI-powered systems for clients" is not a portfolio. "Migrated 309,000 contacts from a fragmented stack into a single CRM with a 93% AI classification rate" is. If the language is consistently vague, the work probably is too.
2. Hourly rate as the primary pricing model for builds. A serious Claude Code engineer prices builds fixed because they're confident in scope and speed. Hourly billing for a build is a tell that they're hedging on uncertainty about how long the work will take.
3. No questions about your operation. If they jump straight to a quote without asking what tools you currently run, what's broken, and what you've tried, they're not selling a build. They're selling a template.
4. "I can do anything." The honest answer is "here's what I'm best at, here's what I won't take." Anyone who says they can build whatever you need probably hasn't built much of anything.
5. Prices that look like AI startup math. A real engagement isn't $500 because Claude is fast. The engineer's time is the cost. Tools amortise across many clients. Prices that seem suspiciously low usually mean spec work, junior work, or a bait-and-switch.
Specific questions to ask on a discovery call
1. Walk me through a recent production system you shipped. What broke after launch and how did you handle it?
2. What's a project you turned down and why?
3. Show me a piece of writing you've published about how you work.
4. How do you scope to prevent creep mid-build?
5. What does your definition of done look like at the end of each week?
6. When is Claude Code the wrong tool?
7. How do you handle the handover at the end of an engagement?
The single highest-signal question: "Tell me about a time the AI was wrong in production. What did the system do, and what did you change after?" Anyone who can answer that has shipped real work. Anyone who can't, hasn't.
What a fair price looks like
Engagements at the senior end start around $6,000 for a small bespoke build. Standard multi-week engagements run $12,000 to $18,000. Multi-location rollouts and complex AI agent builds scale from there. Retainers post-build typically run $2,000 to $3,200 a month.
Below the $6K floor, you're either getting junior work or template work. Above $25K for a single-location single-engagement build, you're paying agency margin without getting agency capacity.
The right buyer for a senior Claude Code engineer isn't the cheapest option. It's the one whose price matches their methodology and whose track record matches their pitch.
The bigger picture
The market for Claude Code experts is going to commoditise. Anyone with a course completion certificate will list themselves in 2027. The differentiator already isn't the tool. It's the operations expertise, the production methodology, and the writing that proves both.
Hire for the operator. The tool is incidental.
Frequently Asked Questions
Running on a stack that grew by accident?
Tools added one at a time, never architected together. That's the problem I solve. Book 45 minutes and I'll map what moves, what stays, and what makes sense for your operation.
Book a Discovery Call