4 August 2026
Capture should be dumb. Review is where the thinking happens.
Most productivity apps try to organise your thought the moment you have it. That's the worst possible time. Notes from building TetherLog.
Views —
Loading…
4 August 2026
Most productivity apps try to organise your thought the moment you have it. That's the worst possible time. Notes from building TetherLog.
Loading…
Most productivity apps ask you to categorise a thought the second you have it. Pick a project, set a priority, guess a due date, answer the chatbot's follow-up question. For an ADHD brain mid-task (mine's AuDHD, to be precise) that's the exact moment you have the least capacity to do any of that.
So the thought gets half-parked, the task gets derailed, and you've lost both. I've closed more todo apps mid-onboarding than I can count, because the fourth screen wanted a due date for a thought I hadn't even finished having.
I'm building TetherLog around the opposite bet: capture and review are not the same job, and putting them in the same interaction is what breaks the whole thing.
The core move isn't new. It's a distraction sheet, the tactic a lot of ADHD (AuDHD) workflow advice lands on eventually: keep a physical sheet next to whatever you're doing, and the second something tries to pull you off-task, scribble it down in two seconds flat and go straight back to what you were doing. No sorting, no folder, just get it out of your head fast enough that it stops shouting at you.
The trick works. The paper doesn't. You can't carry a sheet of A4 through a commute, a client call, and three context switches without it ending up as a ball in the bottom of a bag, and every todo app I tried wanted the sorting done at the exact moment I was trying to avoid doing it.
There was a second motive behind the build too. Review is the part I personally dodge, so a system that depended on my own discipline to show up was doomed the same way the paper was. I wanted the patterns pulled out for me instead of trusted to memory, so I could see my own habits in data instead of vibes: what gets parked and dropped, what actually gets acted on, whether review even happens most weeks. TetherLog is the distraction sheet with pockets, a phone, and a mirror built in.
When something surfaces mid-flow (an idea, a task I forgot, a line for this exact post), I have maybe ten seconds of runway before it's gone or before it costs me the thing I was already doing. If the app asks me one clarifying question in that window, I lose both threads.
The fix isn't a smarter capture box. It's a dumber one.
TetherLog's first rule, straight from the spec: capture is dumb and fast: no AI analysis, no questions, no triage. One tap, text or voice, park it, gone. Tags are optional and never required. If capture is hard, the product already failed.
Voice works the same way. It's not a coaching session: no "did you mean," no live summarising. It's input plumbing. You babble, it transcribes, the mess ships with the capture. Cleaning it up is not this layer's job.
All the intelligence I skipped at capture time shows up later, in the evening review: a separate screen, a separate ritual, run only once the active task is already done.
The review agent walks the day's captures and proposes a bucket for each one:
{
captureId: string
bucket: "do" | "later" | "drop" | "wonder" | "write"
reason: string // one line, warm, literal
carryForward: boolean // max one per session
}
I confirm or override each one with a tap. No key means no agent, but it doesn't mean no review. A rule-based fallback still buckets things (tag now → do, same thought parked three times → drop or do with a note that you've been here before). AI is an enhancement on top of a product that already works without it.
From there, "hands" turn do items into something that actually leaves the app: clipboard markdown, an .ics download, a pre-filled mailto:, a JSON backup. Plain browser APIs, not hosted integrations. The key, when there is one, stays in your browser. TetherLog never sees it.
It's the same shape as any ingest pipeline: write fast now, do the expensive work later, off the hot path.
The review screen also has to stay a calm room, not a guilt trip. No streaks, no red overdue badges, no "you missed yesterday." An untriaged capture from three days ago isn't a failure state. It's a parked thought, still waiting, exactly where you left it.
The moment a review screen starts punishing the backlog, people stop opening it, and the backlog wins by default.
Honest state, not the pitch version:
Built: capture with local persistence, voice (quick parks and long babble sessions, both with transcript and stored audio), the evening review loop with real agent triage and a working no-key fallback, a deterministic pattern dashboard, weekly digest, and the core hands (clipboard, calendar export, share, email, backup).
Not yet: the writing lane, a dedicated write bucket so blog seeds like this post get organised instead of just parked, the one-tap phone install path, and a service-worker nudge for review time (gentle, not a notification shouting at me to do my chores).
That writing lane is the reason this post exists in the first place. Half of it started as a voice dump I hadn't organised. That's not a coincidence. It's the whole thesis working on its own author.
TetherLog is public on GitHub, client-side only, MIT. Clone it, run it, poke at the review logic; there's no account to make and no server holding your data, so there's genuinely nothing stopping you trying it in the next five minutes. If you're building anything for a brain that abandons tasks under friction, the two rules that matter are: make capture embarrassingly dumb, and never let review guilt-trip you for the backlog it made possible.