The Quest for the Perfect AI Writing Tool
I didn’t set out to build a writing platform. I set out to stop copy‑pasting my brain across a dozen clever apps. If you have tried to use AI as a serious writing tool then I'm sure you know what I was dealing with: the feeling of being lost in somebody else’s temple, pulling levers you didn’t design.
The Copy‑Paste Trail
It began with the shiny things—ai that promised flow, clarity, and speed. A new technology had arrived that was incredible at aiding my writing, but it was tedious to work with. I’d draft in one place, paste into another, bounce to a third for style, then back again to fix what the first tool broke. The words moved; paragraphs disappeared and reconfigured like stone blocks shifting in a trap‑filled corridor. Context slipped through the cracks as I had to keep explaining to each new chat instance what I was trying to achieve.
After a while, I didn’t trust what came back to me. If I didn't have a heavy hand in the creation process, the output was clearly slop, and most of these interactions were causing the llm to rewrite portions of my documents that I never wanted to change. It was difficult to keep an eye on what was changing with each iteration. The tooling had a tendency to just start writing all by itself and that is not what I wanted. I want to be the writer and I want a tool that helps me, not a ghost writer lurking behind the walls.
But then new functionality started to change all of that. Artifacts were released, and I began to follow the trail, sensing I was getting closer—like spotting a series of weathered markers leading deeper into the ruins.
The Artifact Cache
I had ventured into the era of apps and artifacts—little containers for ideas that felt more structured than loose prompts. It was better: there was a central place that I could focus my attention with the llm and begin to build out a document over time. These artifacts were even versioned so I could review what had been altered more easily, like flipping through earlier entries in a field notebook.
But the core problem lingered. I still felt removed from the process. I was continually asking the llm to edit the artifact, even for mundane little edits that I wished I could make myself. The process was slow and it was still difficult to ensure that the output wasn't AI-generated slop
and properly shared my views in my voice. I had a cache of relics, but no real grip on how they were being carved.
Agents in the Editor
Meanwhile, my best work happened in a coding IDE. There, the agents worked with me. I could have Claude Code make edits to the code alongside me. I could see what changes it was making and I had clear history of them. Coding agents have evolved to the point that they are almost more enjoyable to use than writing code. In that world, every change is tracked with diffs, commits, and a detailed history.
So I tried a sideways move: treat writing like code and use coding agents to help me write. It was surprisingly close, and if I was writing a technical piece about a code change it worked very well to be able to interrogate the change as it wrote. But the fit was off. These agents were built for compilers, not the nuance of human language. They were fantastic at quickly locating, analyzing, and fixing bugs, or generating new features that matched the flow and style of an application. But when it came to writing, the effort felt lifeless and the collaborative aspects were hard to manage. The agents felt like guides trained for a different continent: brilliant with machinery, clumsy with meaning. This was the map, though. If I could follow this lead to use agents that interact with my documents, maybe I could find the glorious AI writing future that the marketing hype had promised.
Forge the Tool
Scriptum started as an attempt to bring the IDE mindset to content: treat ideas like code, use agents as collaborators, and make progress measurable. If a system could test a function, it could test a draft’s intent, coverage, and claims—differently, but still concretely. I wasn’t hunting for a magic artifact anymore. I was ready to forge my own.
Before I wrote a line of code, I had a map. I had learned how to work with agents and how to work with tools to edit files. All I needed to do was follow the guides that I had seen from the tools I used for coding and the treasure would follow!
The map had worked and now this was the key I needed to help unlock the treasure I was looking for.
The Treasure Chamber
End of the maze, last click, door up. There it was—the thing I came for. Not luck; rhythm. Plan before prompting. Phase everything. Demand validations. Never skip the loop. The pattern turned chaos into craft, and what I lifted into the light was finally mine: a writing engine that keeps my voice, shows its work, and works with me, not over me.
I didn’t just ship a tool—I unearthed it. I brushed off the dust, checked it for traps, and kept only what I trusted. And yeah, it’s sick. That’s how Scriptum runs—and it’s how I plan to build from here on.
Down the Wrong Aisle
Along the way I thought there must be others who have built exactly what I'm looking for. I tried the shelves of the app stores and found fast replies, smart triage, and tidy keyboards, but nothing that focused on researching and delivering a single document. These tools seem more focused on shooting off a quick email that you are too lazy to write. I really wanted a tool that was focused on developing a document and providing rich context. Long ideas need room, memory, and taste. I tested what others built, but nothing followed my path. Nothing had walked the same maze or survived the same traps.
The magic was in building something that matched my taste—and earned my trust.
Aftermath: The Taste of Your Tools
We’ve entered the age of AI‑written software—systems that can draft features as easily as sentences. That’s power, but it’s also responsibility. Tools now inherit your taste, or else they impose theirs. If you care about how your work thinks, you have to teach your tools what “good” means and wire that into the process, like carving your own markings into the tools you carry into the field.
For me, Scriptum is an example of our future—software that adapts directly to my own needs and desires. Your version might look different. That’s the point. Don’t wait for the perfect app aisle. In this future you have the ability to build exactly what you need with little more than a description of the idea. The software will be perfect for you because it was made by you. In the end, the only map that matters is the one you’re willing to follow.