What we found had nothing to do with AI.

01 - The question

Everything changed except the work itself.

Sixth semester there sends every student into industry full-time. Ours landed in the months generative AI was rewriting how software got built.

So we spent them on models, agents and automation, stuck on one question: if this is what the tools can do, why is nothing inside real companies moving?

02 - The startups

The answer was always one desk away.

Saransh's first two jobs were early-stage - one fintech, one consumer software. Ask why anything was built the way it was and you didn't search. You found the person.

Lehar hit it from the other side, in software. Years of meetings, client calls and delivery cycles, none of it compounding. New hires spent a month re-deciding what the company already knew.

Everyone owned a fragment. Nobody owned the whole. Easy to write off as a startup tax.

03 - The plant

We assumed scale had already solved this.

Then Saransh's internship landed in heavy manufacturing - one of the largest plants in the country. We expected sensors, dashboards, AI in the loop. Same problem, just bigger.

Every hard question walked back to the same few operators. Twenty-five, thirty-five years on one machine. They could hear a failure coming before an instrument caught it. None of it written anywhere.

And it never crossed a corridor. Production, maintenance, quality, planning - each learning daily, nothing pooling. Every retirement was a deletion.

Decades of data. Decades of expertise. Nothing joining them.

04 - The thesis

Most companies think they have an AI problem. They have a memory problem.

An agent is only as good as what it can recall. With nothing to recall, it can't see the decision already made, can't reach across a department, can't keep what walks out the door. It becomes one more window a human has to explain things to.

What's missing isn't a better model. It's the brain every company assumes it already has.

05 - The build

So we started building what should already exist.

OXYGN plugs into the systems a company already runs and turns the scatter into one live map of how the place works. Three rules:

1. Rule 1

   One layer, every system

   Knowledge locked in tools that can't see each other helps nobody.

2. Rule 2

   Ask for what was never written

   Indexing documents is easy. The expertise that runs a company was never a document.

3. Rule 3

   Agents that know the place

   Not chatbots. Coworkers - each held to what its user is cleared to see.

Every answer shows its source, or admits it doesn't know.

06 - The invitation

The knowledge is still in the building. For now.

We didn't get here from a report. We got here from four companies, two of us, and the same wall each time. Twenty people or twenty thousand - you are losing knowledge this week.

If your company runs on a few irreplaceable heads, we're building this for you. Come shape it.

- Saransh & Lehar, founders