I built MyText — 161 text manipulation tools — in an afternoon. I had King of the Hill playing on the other screen. That is not a brag about speed. It is a statement about what software development looks like now.
What vibe coding actually means in practice#
Most people outside of software have not heard the term “vibe coding” yet, so here is what it means: instead of writing code by hand, or even asking an AI for help line by line, you describe what you want at a high level and let the AI build it. Claude Code has a setting called YOLO mode — yes, that is the actual name — where it executes autonomously, making decisions and writing code without stopping for approval at every step.
My workflow for MyText went like this. I started with a frustration: I kept pasting text into random websites to do simple operations — formatting JSON, decoding JWTs, diffing two documents, counting words, converting CSV to JSON. Every one of those sites was plastered with ads or wanted my data on their servers. I wanted a clean, fast Swiss army knife for text that I controlled.
I used ChatGPT to help me write an extensive prompt — brainstorming the most common text manipulation operations, adding my own favorites, structuring the request. Then I handed that prompt to Claude Code and let it build. I worked in batches of twenty tools at a time. The quality stayed consistent throughout — the approach scaled without diminishing returns. The collaboration was genuine: I gave high-level direction, the AI proposed structure and implementation, I refined and curated.
The result was 161 tools across thirteen categories, from format conversion and encoding to diff checking and text generation. All running client-side in the browser. No server processing, no accounts, no tracking.
How I used two AI systems together to build faster#
Something I did not plan but fell into naturally: using two different AI systems for different parts of the process. ChatGPT is strong at brainstorming, expanding ideas, and refining prompts. Claude Code is strong at autonomous code generation, understanding project context, and executing across large codebases. Using ChatGPT to think and Claude Code to build gave me a workflow where each system operated in its strength zone.
This is not a product endorsement — it is a practical observation about how AI-assisted development works best right now. The tools have different personalities and different strengths. Using them together, the way you might use different power tools for different parts of a woodworking project, produced better results than relying on either one alone.
I built MyText during a deliberate sprint after Christmas 2025, when Anthropic doubled Claude Code credits. It was part of a conscious push to ship multiple browser-based tool platforms — PDF Pony, GIS Tools, ImageNurse — in rapid succession. MyText was the fastest of them all.
The bottleneck has permanently shifted#
Here is what I actually learned from building 161 tools in an afternoon. It was not about the tools themselves. It was about what the experience revealed about where the bottleneck in software development now lives.
For my entire career, the bottleneck was implementation. You had an idea, and then you spent days, weeks, or months turning it into working software. The gap between concept and execution was wide, and crossing it required sustained effort, focus, and tolerance for tedium. Most side projects died in that gap — not because the ideas were bad, but because the implementation grind wore you down before you finished.
That bottleneck is gone. Implementation is now cheap and fast. The new bottleneck is deciding what to build and why — having the instinct to recognize that a domain is worth covering, the taste to shape something coherent out of a sprawling list of possibilities, and the judgment to make the right architectural calls on privacy and performance. The AI can generate the tools. The human decides whether the whole thing is worth building, and whether it hangs together as a product someone would actually use.
What becomes shippable when tedium disappears#
After building MyText, I started going through old GitHub repos and forgotten laptop drives. Years of half-finished projects — ideas that were genuinely good but that I abandoned at 60% or 80% because I got bored, got distracted, or stalled on the unsexy parts: servers, deployment, the last twenty percent of polish.
Those projects look different now. The question I kept asking myself was one I had not considered before: how tedious and unsexy can an idea be and still make it over the finish line with AI? The answer turned out to be: pretty tedious. Pretty unsexy. Yet now worth finishing.
A word counter is not exciting software. Neither is a CSV to JSON converter or a line sorter or a whitespace normalizer. These are boring, utilitarian tools. Before AI-assisted development, building 161 of them would have been mind-numbing work that no solo developer would voluntarily commit to. Now I can build them in an afternoon, package them into something genuinely useful, and move on to the next idea.
The implication for every developer with a graveyard of abandoned projects is worth thinking about seriously. The ideas you gave up on might not have been too ambitious. They might have just been too tedious for the tools you had at the time.
The tools I keep reaching for#
Out of 161 tools, here are the ones I personally use most often. The JSON pretty printer and JSON linter — because malformed JSON is a daily irritation. The diff checker — for comparing config files, API responses, and document versions. The JWT decoder — because I work with authentication systems regularly and decoding tokens should not require a third-party website. The regex tester — for the same reason every developer reaches for one. The tiktoken estimator — because when you work with LLM APIs constantly, knowing your token count before you send a request saves time and money. The Markdown editor — because sometimes you just want a clean split-pane Markdown environment without opening an IDE.
These are not glamorous tools. They are the kind of thing you reach for ten times a week without thinking about it. Building all of them in one place, running locally, with no tracking — that is what makes the collection useful. Not any individual tool, but the fact that they are all there when you need them.
MyText works offline and processes everything in the browser.
