A brainstorm, on the record
Invited a friend into the project — and he caught fire even more than me. We sat down to prepare the accelerator application and recorded the whole brainstorm straight into the journal: target audience, integrations, and the biggest problem — the input interface.
By Georg Malahov
A week and a half ago I invited a friend to join me in applying to the accelerator. Since then he's been calling, texting, wanting to meet — so active that it seems I got him hooked both on the project and on the application itself. It's an unfamiliar and pleasant feeling: until now the journal was a one-man story, and now there's a person who nudges me, not the other way around.
Today we sat down together and held a brainstorm. The best part — we recorded it straight into the journal: hit record, talked for an hour, and the bot transcribed the whole thing itself. The product documenting its own birth — you couldn't invent a better case of eating your own dog food.
What we argued about. The first question — who is this for? The project was born as a personal assistant: you speak your notes, they get structured, you can search them and ask questions. But there's also a team scenario: record a meeting, get a summary, break it into tasks, send them out to participants — who does what and by when. Deepgram, the transcription service I use, has a dialog mode — it tells speakers apart by voice. For the team scenario that would be exactly the right thing; I just don't have that mode enabled yet. My friend kept pulling me toward the team side, and I kept digging in: sure, you can split the difference, but you have to pick one.
The second question — integrations. A former colleague told me flat out he'll start using it the moment there's a calendar integration. We discussed Zapier — a service that connects thousands of other services to each other; in theory one such integration covers email, calendar, and messengers at once. Another idea — photo recognition: send a photo of a schedule or a sign, the bot extracts the text and stores it next to the note, and later you can search on it.
But my main takeaway is different, and it's not about features. The biggest problem is the input interface. It doesn't matter whether it's a personal assistant or a team one: entering things must be effortless, ideally in the background, like a real assistant. The classic case is the car: you're driving, dictating your thoughts, then someone calls — and no phone ever brings you back into the recording afterwards. Handling interruptions, working in the background, voice activation — that's where it gets decided whether people will use this or not.
And one more decision is ripening: Telegram cuts the audience down hard. Nobody will install a separate messenger just to try your product. So the application needs a standalone web client that a person reaches through an ordinary link.
The application has to go out by the end of February. Two weeks to turn a bot built for myself into something I'm not embarrassed to show to outsiders.