Claude Code Changed My Workflow
I built a Fish Cost Calculator in a weekend that would have taken me two weeks. Here is what Claude Code actually does well and where I still need to think for myself.
The Solo Builder Problem
When you are a one-person operation building software alongside running a seafood business, every hour of development time competes with hours spent on logistics, sales, and compliance. I do not have a team of engineers. I do not have sprints. I have a whiteboard of problems and whatever time I can carve out between fish deliveries and regulatory paperwork.
That constraint forces a question most developers at bigger companies never have to ask: is this feature worth the time it takes me to build it? For years, the answer was often no. Good ideas died on the whiteboard because I could not justify the hours.
Claude Code changed that calculus.
What Claude Code Actually Is
For anyone who has not used it: Claude Code is Anthropic's CLI tool that puts an AI agent directly in your terminal. It reads your codebase, understands your project structure, and can write, edit, and debug code alongside you. It is not an autocomplete engine or a chatbot that generates snippets you paste in. It operates on your actual files, runs your actual commands, and works within the context of your actual project.
I started using it about six months ago. The learning curve was less about the tool and more about learning when to use it versus when to do the work myself.
The Fish Cost Calculator
The clearest example is the Fish Cost Calculator I built for Pacific Cloud Seafoods. The tool takes raw fish purchase data — species, weight, price per pound, processing yield percentages — and calculates true per-unit costs after accounting for shrink, labor, packaging, and shipping. It sounds simple. It is not.
The calculation logic has to handle variable yield rates across species (a whole sockeye salmon yields differently than a whole halibut), tiered shipping costs based on weight and distance, and seasonal pricing that changes weekly. There is also a frontend that needs to be usable by someone standing in a processing facility with wet hands and a phone screen.
I built the core calculation engine with Claude Code in a weekend. The process looked like this: I described the business logic in plain language, Claude Code scaffolded the functions, I reviewed every line to make sure the math was right, then iterated on edge cases. The yield calculations for different species and product forms — fillets versus portions versus whole fish — would have taken me days to test exhaustively. Claude Code generated the test cases faster than I could have written them.
Total time: about twelve hours of focused work. My estimate for doing it solo without AI assistance: two weeks of evenings and weekends.
Where It Excels
Boilerplate and scaffolding. Setting up a new Next.js page with the right imports, layout structure, and TypeScript types is tedious but necessary. Claude Code handles it in seconds and gets the project conventions right because it reads the existing codebase first.
Test generation. This is probably the highest-value use case for a solo builder. I am honest with myself: I under-test my code when working alone because writing tests feels like it slows me down. Claude Code removes that excuse. I describe what should be tested and it writes comprehensive test suites that catch edge cases I would have missed.
Debugging across files. When a bug involves the interaction between three or four components, Claude Code can hold all of them in context simultaneously. I had a production issue where my knowledge graph component was rendering correctly in development but breaking on the static export. Claude Code traced the issue across the dynamic import wrapper, the graph data file, and the Next.js config in about two minutes. That kind of cross-file debugging used to eat entire evenings.
Reading unfamiliar code. When I integrate with third-party APIs — NOAA data feeds, shipping rate calculators — Claude Code can read their documentation and existing code samples, then write integration code that follows my project's patterns. It saves the most painful kind of development time: the hours spent reading docs for a library you will use once.
Where Human Judgment Still Matters
Business logic validation. Claude Code can implement a yield calculation, but it cannot tell you whether 62% yield on sockeye fillets is realistic. That comes from standing in a processing plant and watching fish get cut. Every calculation Claude Code writes needs to be validated against real-world numbers, and I have caught errors where the generated code used plausible-sounding but incorrect assumptions.
Architecture decisions. Should this be a separate microservice or a function within the existing app? Should I use server-side rendering or static generation for this page? Claude Code will do whatever you tell it to, which means it will happily build the wrong thing if you make the wrong architectural call. The tool amplifies your decisions — good and bad.
User experience. AI can generate a functional interface, but it does not know that the person using this tool is wearing fish-processing gloves and squinting at a phone in a 38-degree cooler. Those constraints come from lived experience, not training data.
Knowing what to build. The hardest part of solo development was never the coding. It was deciding which of the twenty things on my whiteboard would actually move the business forward. Claude Code makes building faster, but it does not make prioritization easier. If anything, the speed makes it tempting to build everything, which is its own trap.
The Honest Assessment
Claude Code has roughly doubled my output as a solo developer. That is not marketing language — it is a practical observation based on shipping features over the past six months. Things that I would have deferred or simplified now get built properly because the time cost dropped enough to make them worthwhile.
It is not magic. It does not replace understanding your domain, your users, or your codebase. The developers who will get the most out of tools like this are the ones who already know what they want to build and why. The tool handles the how.
For solo builders and small teams, that is the real value. Not replacing the developer, but removing the tax that makes small-team development feel impossible. The whiteboard still has ideas on it, but fewer of them die there now.
