Handwriting was always problematic for me. My fifth-grade teacher, Mrs. Cloud, placed a high value on well-formed cursive strokes that she flowed smoothly onto the blackboard. At my desk I struggled to copy her examples and failed miserably. In middle school, with no one judging my handwriting, I abandoned cursive in favor of printing my letters which was slower but at least I could read what I wrote.
In college, needing to take notes faster than I could print them, I forced myself to relearn cursive. In the 1970s my portable device wasn’t a laptop computer, it was an electric typewriter that I used only for final copy. I composed in longhand on yellow legal pads. By the early 1980s, when it finally became possible to compose on a computer, I thought I’d left handwriting behind forever. Take that, Mrs. Cloud! No more clumsy scribbling with pen and paper! Or so I thought, until the keyboard began to take its toll.
My struggles with RSI began in the waning days of BYTE magazine. I’d been working obsessively for months to complete a subscriber version of byte.com. Just as I was ready to launch it, CMP bought BYTE from McGraw-Hill only to shut us down immediately. I went home, began writing Practical Internet Groupware, and soon realized I’d done real damage to my hands and wrists. So for the rest of that summer I wrote longhand on a series of yellow legal pads.
This was ironic because my beloved Captain Kirk keyboard, later memorialized in the New York Times, was the most ergonomic typing setup there’s ever been before or since.
But even those keystrokes got to be too much. When I advocated for blogging as a mode of communication that optimizes for the amount of awareness and influence that each keystroke can possibly yield, the subtext was relief for my aching hands.
Voice input was always the dream. Periodically I would try the latest version of Dragon Naturally Speaking but it never worked fluently for prose and was hopeless for code. Until fairly recently, RSI-challenged programmers went to extraordinary lengths to code by voice. In this 2013 video Tavis Rudd demoed a method that required a huge specialized vocabulary to express commands, functions, variables, punctuation, and cursor movement. Where there’s a will there’s a way, but I knew that wasn’t for me.
A few months ago, as I began developing Bram, I realized it was time to give voice recognition another try. If you’ve used any form of it recently you’ve noticed the improvement as the rising tide of AI lifts all boats. It’s gotten way easier to dictate prose reliably. And now, suddenly, that’s also a way to produce code. You don’t have to express commands, functions, variables, and punctuation, you describe outcomes and monitor agents that do most of the writing and editing. I connected Bram to a Whisper server and the results have been dramatic. Here’s what went into last night’s v0.2.22.
– Self-heal stuck “delete pending” session rows
– Add a New session button to the Sessions page
– Render Supabase execute_sql as pretty SQL in, table out
– Add a Skills launcher to the agent pane (#221)
– Default continueLast to on so restarts resume the session
– Instrument session rotation so it names its own cause
– Force a full transcript fetch on window-miss to survive session rotation
– Observe-only: flag user-interrupt-after-permission turn ends
– Fix send-ledger false-strand across a session switch
– Render Codex exec-wrapped apply_patch as a diff
– Observe-only: flag send-ledger false-strands across a session rollover
– Use a browser-safe HTTP URL in the Target app info dialog
– Toast when a Push auto-closes issues
– Auto-close issues on push; remove agent close route (security H5, #118)
– Revert “Host-authorize issue close side effects (security H5, #118)”
– Host-authorize issue close side effects (security H5, #118)
– Widen the H4 authorization TTL to fit implementation time
– Parse Codex unified-exec (custom_tool_call name=exec) tool cards
All this required almost no typing, I just used my voice to direct Claude Code and Codex to do the research, coding, and testing. Take that, Mrs. Cloud! No more clumsy scribbling, and no more typing either. Finally I can build software by just talking to the computer. It really is a dream come true.
