What is Claude Code? The €600 Freelancer Quote I Killed in

By Lazar Milicevic · Published June 12, 2026 · 7 min read

Last Tuesday I almost wired €600 to a freelancer to add multi-currency support to a client's invoicing app. Instead I opened a terminal, typed one paragraph, and shipped the feature in 38 minutes for about $4 in API credits. This isn't a story about a cheap freelancer being undercut — it's about what Claude Code actually is, and why the name undersells it for business owners who already own custom software.

The €600 quote that triggered this

The client runs a small invoicing platform for Serbian businesses. Until last week it only handled local currency (RSD). A customer asked the reasonable question: can I bill my German client in euros and my American client in dollars?

The freelancer scoped it honestly:

Schema change (add currency column to invoices)
API endpoint update (accept + validate currency)
PDF template adjustment (symbol + ISO code rendering)
Testing + deployment

Quote: ~€600, two working days. Fair price for the actual work. The problem isn't the freelancer's rate — it's that the business owner had four other small requests stacked behind this one. €600 × 5 = €3,000 just to keep the software matching how the business actually operates. That math collapses fast for a small business.

Claude Code is not a chat window — that's the category shift

Most business owners hear "Claude Code" and picture a chat window where developers paste code and get suggestions back. That's the old loop. The chat window puts you in the loop on every line.

Claude Code is different. It's an agent that:

Lives inside your terminal
Reads your entire repository (files, structure, dependencies)
Edits files directly on disk
Runs commands (migrations, tests, builds) and reads the output
Adjusts based on what it sees

You describe the business rule. It handles the translation into code. The agent puts you in the loop on the decision, not the line. That's the only shift that matters for a non-developer owner.

The exact paragraph I typed

I opened the project folder, launched Claude Code, and typed this. No code, no file names, no SQL:

Add support for euro and dollar invoicing alongside the existing
currency. Update the database schema with a currency field on
invoices. Adjust the creation endpoint to accept and validate the
currency. Update the PDF template so the symbol and code render
correctly. Keep existing invoices working without migration
breakage.

That's it. Written the way I'd explain it to a junior employee who already knows the codebase. Notice what's not in there: no table names, no framework references, no library calls. Just the business rule and one explicit constraint about backwards compatibility.

What it actually did before writing a single character of code

Opened the invoice model file
Opened the controller handling invoice creation
Opened the PDF generator template
Mapped the existing currency assumptions across all three
Proposed a plan: 3 files to edit, 1 migration to write, 1 template change

Then it asked me to confirm. I said yes.

The 38-minute breakdown

This is what actually happened, in order:

00:00  Launch claude in project root
00:02  Paste the paragraph above
00:04  Agent reads repo, returns plan + diff preview
00:06  I approve plan
00:08  Migration file written
00:09  Migration applied to local DB → "OK, no errors"
00:14  API endpoint edited + validation added
00:21  PDF template updated with symbol/code mapping
00:24  Agent asks: run the test suite?
00:25  I say yes
00:31  Tests pass (existing + new)
00:34  I manually create a test invoice in EUR via admin panel
00:36  PDF renders with € symbol + "EUR" code correctly
00:38  Done. Commit.

Cost on the API meter: $4.12. Elapsed: 38 minutes. The €600 / 2-day quote wasn't overpriced — the freelancer would have done exactly the same work. What got removed wasn't labor. It was the translation tax: the back-and-forth emails, the scheduling, the rebuilding of context every time we picked the project up again.

The real skill: describing business rules clearly

Here's the part nobody tells business owners. You don't need to learn to code to use this. You need to learn to describe your business rules clearly — and you've been doing that for years with employees, freelancers, and accountants.

Compare a vague prompt vs. a clear one:

BAD:  "Add multi-currency to invoices"

GOOD: "Add support for euro and dollar invoicing alongside the
       existing currency. Validate currency on the creation
       endpoint. Render the correct symbol in the PDF. Don't
       break existing invoices — no forced migration of old
       records."

The good version isn't more technical. It's more specific about the business constraints: which currencies, where the validation lives in the user's experience, and what must keep working. The terminal stops being a developer tool and becomes your admin panel for the parts of your software the actual admin panel doesn't expose.

Things you can describe instead of coding

A new tax rule (rate, threshold, which customer segment)
How late fees calculate (grace period, percentage, compounding rules)
A new report format your accountant needs (columns, grouping, date range)
A new invoice status transition (when does "sent" become "overdue")

The honest limits

I'm not selling fire-and-forget. There are real failure modes:

Vague descriptions produce wrong assumptions. If you say "fix the tax calculation" without saying which tax, which country, which customer type, you'll get a confident wrong answer.
It sometimes picks a more complicated solution than needed. Reviewing the plan before approving is not optional.
You have to test the result. Tests passing ≠ feature correct. I manually created a EUR invoice and opened the PDF. That step is yours.
Large refactors are riskier than small features. The 38-minute success was a well-scoped change to three files. A "redesign the whole billing flow" prompt is a different conversation.

Think of it as a fast, literal contractor who never sleeps and charges by the minute. Brilliant at small, well-defined tasks. Dangerous if you skip the review step.

Why bizflowai.io helps with this

For clients who already own custom software but don't want to babysit a terminal, this is exactly the kind of work bizflowai.io handles — running the agent loop on their codebase, translating their business requests into clean prompts, reviewing the proposed plans, and shipping the small features that used to sit in a freelancer backlog for weeks. The owner stays in the loop on decisions; the translation tax disappears.

Frequently asked questions

What is Claude Code?

Claude Code is an AI agent that runs inside your terminal, not a chat window for pasting code. It reads your entire repository, edits files directly on disk, runs commands, sees the output, and adjusts. You describe a business rule in plain language, and it handles translating that rule into code changes, asking you to approve a plan before executing.

How do I add a new feature to my software using Claude Code?

Open your project folder, launch Claude Code, and type a paragraph describing the business rule in plain English—no code or file names needed. The agent reads your repo, proposes a plan listing files to edit and migrations to write, and asks you to confirm. After approval, it writes the code, runs migrations, and can execute your test suite.

Do I need to know how to code to use Claude Code?

No. You need to describe your business rules clearly, the same way you'd explain them to a junior employee or freelancer. The agent reads your existing code, understands the logic, and proposes changes. You approve or reject the plan. The skill required is clear written description of rules, not programming syntax.

When should I use an AI coding agent vs hiring a freelancer?

For small, well-scoped feature requests on software you already own, an agent is dramatically cheaper. In one real example, adding multi-currency invoicing cost roughly 600 euros and two days with a freelancer, versus about 4 dollars and 38 minutes with Claude Code. Freelancers still make sense for large, ambiguous, or architecturally risky work that needs deep human judgment.

What are the limits of using Claude Code for business software changes?

Claude Code can make wrong assumptions about business logic if your description is vague, and it sometimes picks more complicated solutions than needed. You must review the proposed plan before approving and test the result afterward. It is not fire-and-forget—treat it like a fast, literal contractor that needs clear instructions and verification of its output.

Want more like this?

I publish practical AI automation, GenAI engineering, and faceless content workflows on YouTube every week.

Subscribe to bizflowai.io on YouTube — never miss a new tutorial.

Planning an AI automation project or need a second opinion on your architecture?

Connect with me on LinkedIn — Lazar Milicevic, GenAI Engineer & bizflowai.io Founder.