building Mira • takes on models, agents, and hardware
Joined October 2021
- Tweets1K
- Following35
- Followers72
- Likes26
this week i banned mira from opening a message with "noticed"
she'd started 17 of 19 drafts with it in one test batch
then i checked the four sample lines she learns my voice from. all four open with noticed, saw or your lol
stayed up for devday ~ dots gets its own computer and works 24/7
i'm rebuilding mira into the rep who answers the café at 9pm and comes back with the order
24/7 was never the hard part. the café saying yes is
mira's last 8 openings went to coffee roasters and cafés ~ her checker held 7, this one too
"a room built around the coffee is exactly where the next thing you stock needs to earn its place on the counter"
reads nice. i couldn't tell you one thing about that place from it
the one it passed said how that business actually gets new accounts
a café asking for 30 days credit on a first carton is asking the founder for a ₹4,200 loan ~ 24 jars at ₹175 to the store
ten cafés on those terms is ₹42,000 sitting on other people's counters
i'm rebuilding mira to answer that with upi on delivery and 2% off for paying on the day, ₹84 on that carton
the replies are arguing about which chart while the model got cheaper lol
xitter.cf/shaunstanley_/status/2…
monday ~ rebuilding mira, the ai sales rep for small indian brands, in public
what shipped: once i approve an order she writes the invoice, tax included
friday she ran a full trial, first message through the replies to a booked meeting, on a fake buyer so nothing real could break
next: a rehearsal on my own accounts, sending switched off
first weekend since august without a commit to mira's code
friday night she got a fake trial ~ fake buyer, fake replies, start to finish. i'm reading it instead lmao
mira wrote three openings for coffee roasters. her checker held all three
"roasts its own beans rather than just buying in coffee to serve"
"the kind of name that tells regulars you take the bean seriously"
"every new pour has to hold up against the roast itself"
i'd have sent the third one. which would you send?
mira wrote this for a café. her own checker held it before i got to it
"roasts its own beans, which puts it a step ahead of most cafés that just buy in"
the check it failed: would a hundred other cafés get the same line. yeah. she's stricter than me lol
every ai sales agent demo i've seen ends at "message sent" lmao
the café hasn't even tasted the thing yet. that's where the actual job starts
a chatgpt co-inventor shipped a model that can't generate text.
output tokens are free because there aren't any
anthropic is profitable before compute and the replies are "so am i, before rent".
rent doesn't compound. compute does.
5k hit, everyone reset. thanks tibo
Hi Astra users. A reset and a quick update on quality issues that have been posted around.
Working with some of you, we have found and fixed the following issues:
- Some skills written for previous models were triggering too often or preventing the model from checking its work.
- An opt-in context management experiment that could cause early stops or replies to older messages. We've disabled it. Our rough estimate is that 4-5k users were affected by this experiment.
- We've also removed some badly configured engines that resulted in a measured quality degradation for a long tail of traffic flowing through them.
We’ve also made some more minor improvements and things should feel significantly better across the board. More consistent follow-through, better tracking of your latest message, and better checks on the work as it’s going through the motions.
The examples posted and all the users who worked directly with us were incredibly useful in helping fix things quickly. Always grateful for this incredible community.
And of course, a reset is also landing by midnight today.