shipped a few things this week.
first, better intent mapping. it sounds plain but theres a lot that goes into figuring out what actually drives people. were still learning from the data we have.
second, and i think the most useful thing we shipped, a technical / non-technical view toggle. managers dont always need to see the jargon. a lot of the time they just need the signal in clean language. so we started stripping the tech overhead and handing things over in plain terms. needs me today, open promises, blockers, status updates. you shouldnt have to infer anything.
third, open promises. inspired by something kunal shah did for himself, but at an organisational level. a place to see what people said they would do and whether its still open.
the deeper thread under all of this is still intent. you really cant fully map someones intent unless you are that person. but the highest leverage point sits right there. the better you can map intent, the better you can predict what comes next. for a developer that intent is basically whats still keeping the job alive. an agent can do almost everything else.
so in a way this whole build is about bridging the gap between agents and the real world. taking data from the stakeholders youre eventually going to replace. its a strange thought but i think its also just where things are headed.
more to come.
this was the first proper demo after about three weeks of building.
the core thesis is pretty simple. most people building in the enterprise memory space are focused on documents and internal ip. i think theyre looking at the wrong asset. a companys biggest assets are its employees. whoever gets fastest at replicating whats actually working with those people is going to build the first real ai organisation.
the bottleneck in most agentic systems right now is how you get agents to work together. if theyre meant to replace human systems they have to work the way humans already work. layers, teams, coordination. so i started hyper niche. software engineering teams that live on slack and github.
one byproduct of going down this path is that tools like slack and teams start looking like temporary infrastructure. their business model runs on keeping people inside the app. that is the opposite of efficiency. if agents are serious about removing work they will eventually route around products built to maximise time spent.
the prototype itself is still low level. we ingest signals from slack and github. theres a champion console that stays with the manager. at the end of the day you can run simulated agent meetings, one agent per employee, to map out what each person needs to do next. in the bigger picture this is the learning phase. the humans are still doing the work. the system is watching how they do it.
we map work conflicts, intent, team heat, tomorrows focus. theres a live graph of people and commits. theres even an unapproved digest so developers can keep some things from being published. small insurance while the system is still learning.
were in the classic chicken and egg. more data makes the product better, but to get more data you need a better product. if any engineering teams want to try it and help us improve, reach out. im just a college kid trying to make this work.
