we've been quiet for a while — focused on building, not on talking about it.
capslock is a small ai lab. we co-create systems with a handful of companies: we get inside how each one actually works, and build the tools it needs, together. el fin is one of them. this is the part where we start opening the notebook on what we've found.
what we actually think about ai
here's the thing we kept proving to ourselves, quietly, for months: the useful part of ai is almost never the part everyone's excited about.
it isn't a giant model thinking for you. it's a set of small, well-wired systems — good semantics, correct timing, real behavioral design underneath. the leverage was never in spending more. the ai behind most of what we run costs about a coffee a week; the value is in wiring it up well, not in burning tokens to look impressive.
and we were wrong about one thing going in. we assumed companies were held back by access — to the right models, the right tools. they weren't. they were held back by fragmentation and friction: knowledge stuck inside teams, work surfacing in the wrong place at the wrong time. the hard part was never the ai. it was everything around it.
╭─────────────────────────────╮ │ ~/capslock — build │░ ├─────────────────────────────┤░ │ caps:on ● building │░ │ │░ │ ███████████████░░░░ 78% │░ │ › wiring systems … │░ ╰─────────────────────────────╯░ ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
the model
so we built for that instead.
one shared core — closer to a shared IT department than a software product — serving a few companies at once. because the foundation is built once and reused, what sits on top can be hyper-personalized: tools shaped to one company, not generic software everyone has to bend to fit. cheap to run, and the IP stays in-house.
and it compounds. something we figure out for one company becomes something the next one gets. nudge, a small system for keeping work moving, was one of the first — and it's already traveled between companies. (here's how that works.)
the leverage was never in spending more. it's in wiring it up well.
from el salvador
we build all of this from el salvador, as a salvadoran company. not because the work has to happen here — honestly, it could happen anywhere — but because this is home, and we'd rather build its future inside it than watch it get built somewhere else and imported back. (more on that.)
the surprise
the most surprising thing we found wasn't technical. it was capslock itself.
we set out building tools. partway through, the thing we were building stopped feeling like a service and started feeling like an idea that needed to exist — and one we got to experiment our way into. the need was real. the benefit was obvious once it was running. and there's a particular pull to building something that, the more you look at it, clearly wants to be built.
why we're opening the notebook
because we think this quieter, earned way of using ai is worth showing — not as claims, but as work. so we'll share what we find here as we find it: what we build, what breaks, what surprises us, what we're still figuring out.
that's what the fieldnotes are. this is the lab, with the door open. we're only getting started.