A hundred euros a month for speed
The structure frontend built in a day, a plan for a rough MVP by the end of the week — and a paid Max subscription to Claude as an investment in development pace.
By Georg Malahov
A good day for Referity — and a good week: at the start of it I beat email delivery from production (the hoster was blocking one port; the mail went through after moving to another), so user notifications finally arrive. And today I finished the frontend part that displays the whole structure: projects, themes, topics — and the overall task list, where each task carries a link to the entity it belongs to. Click it — and you jump into the specific project or theme. What used to be parsed from voice somewhere in the depths of the database can finally be seen with your own eyes.
Three things are on the plan for tomorrow. First — invitations: invite a person into a specific project or theme by email. Second — pass the ID of the entity you're in when uploading audio: a short recording carries too little context for the parsing to reliably tell which project it belongs to — this way the information lands in exactly the project the recording was made from, simpler and more dependable, with auto-detection still in place. The people invited into it will get an update notification. Third, already a bigger step — merging entities: when the parsing has spawned two similar topics, you need to be able to merge them, and it's better not to delete the merged one but to mark where it flowed into. If I manage the first two tomorrow, then this week — very roughly, but still — the MVP closes.
A separate thing worth mentioning is how all of this gets built. Today's structure interface I put together with the help of Claude — the AI assistant from Anthropic, which has a mode for working with code. I tried it — and immediately paid for the Max subscription at €100 a month. That's a noticeable sum for me, but I decided I will unquestionably get my money's worth: it's an investment in the speed of my projects. Subjectively it's in some ways even more interesting than Cursor, the AI code editor I've been using since winter: the way it handles context is clearer — you can see what the model keeps in mind and why.
The arithmetic here is simple. The project lives on the hours left over after the day job and family, and those hours are always fewer than the ideas. A tool that makes each such hour more productive is the most sensible expense there is: today's structure interface went up in a day, and I still had energy left to write down tomorrow's plan.
Next thought from the same series: try moving part of the parsing workflows out of n8n — the automation builder where all the logic lives now — into ordinary code. It probably won't make development faster, but it will make it simpler: typing and tests versus clicking blocks together. I'll check that once the MVP is done.