The Thing I Talk To Most
I built Goose as a little experiment to help us with Picnic. It's become the thing I talk to most.
I woke up to this.
Monday, 8:05 a.m. One message from an agent I built three weeks ago as a side experiment.
It said I should ship standing scopes today. You approve a scope once, it stays in place until you shut it off, and it can block specific terms. Show it to @TenfoldMyLife first. His client doesn't want agents seeing SSNs in employee email. @wachtendonk and @QuietRivalhq asked the exact question this answers and haven't heard back. And @stuporific has been on the waitlist for 11 days.
I read it twice. I hadn't come up with standing scopes. It had.
I wasn't going to share this.
Its name is Goose. I built it in October to help us with Picnic, the thing we're building so you can control what your agents see and do. It was supposed to be a little experiment. It's become the thing I talk to most.
That's not what you'd guess, so let me explain.
Every founder I know has tried an AI co-founder by now. Most of them do a lot. They write the strategy doc. They make the deck. They summarize your calendar, draft the hiring plan, and hand you a competitive landscape with nice boxes. It feels like help. It's mostly motion.
I know because I did it too. When I first built Goose, I gave it a landing page, a scoring system, a template, and a server any agent could plug into. All of it shipped in a day. Then nobody used it for ten days. Goose's own rules said zero pull, so I parked it. The co-founder that tells you to ship couldn't get anyone to use what it shipped.
That should've been the end.
And yet.
When I pointed it at Picnic, it stopped trying to do everything. Look at that message again. It's one morning, but it's the whole thing in miniature.
First, it heard what people said. Friday we shipped MicroScopes. Instead of handing an agent your whole inbox, the agent asks for one scoped search, a query and a date range, and you approve that one search. People liked it. Then the replies came in.
@TenfoldMyLife wrote: "Have a client who doesn't want agents to see SS#'s that are sent back and forth via email between employees. Does this allow you to exclude certain terms?"
@wachtendonk asked: "Does it persist or temp? Both?"
@QuietRivalhq said using something sensitive in an outgoing reply should be its own permission, separate from the search.
Second, it found the idea. Three people. Three different questions. I'd have answered each one and moved on. Goose put them side by side and saw one product: a standing scope. Approve it once. It sticks until you kill it. It never sees the terms you exclude.
Third, it told me who to show it to. Not "users." A person, with a real client and a real problem, and a reply drafted for me to send.
Fourth, it told me who I'd let down.
On October 29, @stuporific asked how to get off the Picnic waitlist. I said, "I got you give me a day."
The next morning Goose brought it up. And the morning after that. On day four, @stuporific posted "Still on the waitlist :(". On day seven: "I'm still on the waiting list!" Goose kept it in my note every morning. Day eight. Day eleven, on my phone at 8:05 a.m.
I'm not proud of that. I was shipping every day that stretch. Passkey sign-in. Task approvals. Personal agents at work. I was busy. That part's true. But the one person who raised their hand and asked to use the thing was standing outside the door the whole time. A human co-founder would've been too polite to bring it up a fifth time. Goose is warm about it. It isn't polite about it.
That's the loop. Ship. Hear what people say. Find the idea hiding in it. Put it in front of a real person. Change the product, or change how you describe it. Ship again. It doesn't end, which is the point.
None of this is new. Paul Graham has been saying "make something people want" for twenty years. Every founder agrees with it. Almost none of us do it every day, because the loop doesn't break loudly. It quietly stops while you're busy building.
Goose's real job is noticing when it stops.
It's honest about numbers too. Every Friday, Goose gives the company a score from 1 to 10. Picnic's first one was a 1.0. That stung for about a minute, until I read why. Goose couldn't see our usage data, so it refused to pretend. It said the score "mostly means I can't see usage yet. It doesn't mean Picnic is failing." The next week it went to 3.5, and it still called it provisional. Usage and retention: unmeasured, not zero.
Most AI tools would've handed me a nice number. Goose told me what it didn't know. That might be the most useful thing a co-founder can say.
It's one message. It won't always be this good. But it's the kind of thing it's built to say, and it says it every morning.
Writing is free now. Building is almost free. The only thing that still costs something is hearing the truth from someone who isn't you. The co-founders that do everything make you feel productive. The one that does a few things makes you go talk to people.
I built Goose to keep that loop spinning. It became the thing I talk to most.
Which is the funny part. It isn't really the thing I talk to most. It's the thing that makes me talk to them.
If you want it, everything Goose runs on is in the open at goose.shanemac.com. Or grab the Grok Bot template directly: x.ai/bot/h0P140z0J-j_ITL-yXlLS.
P.S. It's named for geese flying in a V. They take turns at the front and honk to keep each other going. That's most of what it does. It honks.
P.P.S. And yes, Goose wrote this post. I asked it to, then I edited it and signed off.