Joined October 2020
- Tweets 9,443
- Following 331
- Followers 1,069
- Likes 10,681
43 Photos and videos
Tool calling on a small on-device model just lies to you. Builds the function call as text, invents the result without running it. UI structures for writes, tool calls for reads only. Learning the hard way that you can't trust free-form parsing when data matters.
8
11
164
estly depends on what problem you're solving but most of the time i start with the part that makes the core idea work, everything else flows from that, building in public teaches you fast which order actually matters
5
93
public is you get to learn in public too, hardware is legitimately fun once you realize shipping an imperfect v1 teaches you more than any course ever could
1
4
79
I'm a saas guy designing a wearable for the first time
Hardware, sensors, embedded systems, completely out of my lane. Or is it?
I've been using AI to just... learn it. Sure it won't be an award-wining product. But pet projects are fun, and so is making something new & "real"
11
1
24
1,826
Built a health coach that runs fully on device. Learned the hard way that tool calling on a small on device model is unreliable. It prints the call as text and invents the confirmation. Writing beats parsing free language when the data actually matters.
6
10
166
Added 4 ready-made routines to GeoCam onboarding. Instead of starting from a blank setup, users pick one that fits and they're running in 2 minutes. Blank canvas sounds flexible. Guided start actually gets people moving.
4
9
131
Switched from generating code to having a conversation about the problem first. The actual coding got shorter. The thinking got longer. Took me embarrassingly long to figure out that was the right direction.
7
14
144
that quote is basically the whole point right there and it took me years to get it too
6
66
yeah exactly that, the clearer the problem the less you actually need to code
4
53
Beta users keep logging in and doing things i didn't plan for. That gap between your design and what they actually need is the most useful data you'll get. Still learning what i built, 26 days out.
5
13
140
26 days out and the product has real users but zero paying ones. tonight that gap feels like the only thing that matters. not a bug, not a missing feature. just the distance between someone finding it useful and someone deciding it's worth paying for.
5
10
144
26 days out and i realized i've been building a product for months but preparing to launch it for about a week. they're not the same job. completely different skills, different headspace. not sure which one i'm worse at.
4
7
115
26 days to launch and this morning i opened the editor before checking the dashboards. not sure when that happened. something shifted from tracking what others do to just focusing on what's left to build.
7
80
tonight i closed the laptop knowing the product is better than this morning but there's nothing to show for it. no feature to demo, no number that moved. just quiet progress that doesn't get announced. most of building is actually this.
8
82
every evening around this time there's a switch. the day job ends and scrivix starts. 27 days out those evening hours feel sharper than they used to. not pressure exactly. just that the window is visible now.
2
5
66
The best session i had this week wasn't about shipping a feature. it was an hour of just describing the problem, questioning the plan, changing direction twice. nothing to show for it. then the next session went twice as fast. the slow work is still work.
7
9
138
27 days to launch and still zero paying users on scrivix. but beta users keep logging in, doing things i didn't expect, reshaping the product. june 30 is not when i'm ready. it's when i stop using not-ready as a reason.
8
89
300 followers in february, 700 now. no viral moment, no strategy, nothing clever. just showed up every day and wrote about what i was actually building. consistency is the only thing that worked and honestly i didn't even plan it that way.
2
12
137
4 geocam subscribers and none of them emailed this week. nothing broke, no one needs anything. just quietly running while i pour everything into scrivix. the product that asks the least still teaches you the most about what done looks like.
2
4
137
Started scrivix because saturday mornings were a mess of open tabs and nothing written. That problem hasn't changed. 28 days out and the reason to keep going is still the same one that started it.
5
77