it’s 2am and i’m reading someone’s inbox.
not scanning it. reading it. the way you’d read a friend’s messy desk if they handed you the key and said, tell me what you see. invoices they forgot about. a supplier chasing a late payment. a calendar invite for a meeting they clearly don’t want to go to. i open a blank doc and start writing a briefing, the kind i imagine a very good, very calm chief of staff would hand someone before their morning coffee. three lines. what needs you today. what can wait. what already got handled.
there is nothing automated about this. not one line of it.
the product is called briv. the pitch, if you catch me on a good day, is zero-prompt ai chief of staff for small business owners. connects to your inbox, your calendar, your books. tells you what matters before you ask. it sounds like software. for the first several weeks it was me, a google doc, and a level of attention to a stranger’s business that i have honestly never given my own.
i want to be clear that this wasn’t a clever growth hack dressed up as a story later. it was the only way i knew how to find out if the idea was even true. you cannot design the shape of a good morning briefing by imagining what a busy owner needs. you have to sit inside their inbox and feel what’s actually noise and what’s actually a fire. some of that you only learn by doing the un-automatable version of the job with your own hands, badly, at 2am, for someone who has no idea their chief of staff is a 24 year old with too many tabs open.
so that’s what i did. beta user by beta user. inbox by inbox. writing briefings by hand and sending them out under a name that promised something i hadn’t built yet.
here’s the part i don’t say enough. it worked. not the automation, the deception of it, if we’re calling it what it was. it worked because doing it manually taught me things no dashboard would have surfaced. that a briefing isn’t a summary, it’s a decision someone doesn’t have to make anymore. that the most valuable line in the whole thing was never the biggest number, it was the smallest, weirdest detail. a supplier’s name spelled wrong three emails in a row. a client who always replies at 11pm. the stuff a model would flatten into a category, but a human reading closely would flag as this person’s actual life.
i don’t think you can build zero-prompt anything without first being the prompt yourself. fully, embarrassingly, by hand.
we’re automating it now. slowly, carefully, trying not to lose the thing that made it work in the first place. and i’ll be honest, some days that feels like progress and some days it feels like i’m quietly deleting the best part of the product, the part where a person actually paid attention.
i don’t have a clean ending for this yet. the manual version might have been more honest than whatever we ship. i’m not sure automating something makes it better. sometimes it just makes it scale.