AI Governance 2: Aderit.ai Research Note
Making sense of the data
Research Note: Aderit.ai
Poking holes isn’t a noble instinct. It’s just the one I was issued. I got to do it last week with Darin Ries and Bennie Reddin, the founders of Aderit (aderit.ai). I had more fun than you’d imagine given the topic.
Aderit sits underneath a company’s existing HR systems and pulls them into one shared record of the workforce. Everything reads from it. The founders call the data layer Genome and the agent layer Cortex.
I picked the job description as a test case because it’s simple and messy.
Look closer, and it gets worse
A fractal stays jagged no matter how close you get. The mess is the whole point.
A job description is a fractal. To a recruiter, it’s a sales flyer. To a comp analyst, it’s a category with a pay band. To a manager, it’s small tasks a person can be scored on. To L&D, it’s an inventory of skills. To the worker, it’s the thing they do all day, which looks like none of the above.
Stack them all together, and it still won’t describe the moment-to-moment work.
Ask HR, payroll, and finance what headcount is, and you will get six answers, no agreement, and no fabrication. For thirty years the reflex has been to grind all of it into one clean truth.
That is the mistake.
The good part is the pressure test
They came to get their ideas pressure tested. It’s the polite version of a bar fight.
Darin’s pitch was that AI governance is really a data lineage problem. I pushed back. Governance is thirty-odd control points and sign-offs. Lineage matters, but it isn’t the whole animal.
Bennett didn’t fold.
He tracks eighteen traits of what he calls agentic readiness: not just where data came from, but why, when, and how. I left less sure of my own point. Pressure testing cuts both ways.
Many truths, kept apart
You know the old story. One blind man holds the elephant’s trunk and calls it a snake. Another holds a leg and calls it a tree. Each is right, and neither has the elephant.
The job description is the elephant. Everyone who holds a piece is sure they have created the ultimate JD.
To a recruiter, the job description is a sales flyer. It’s bait, written to attract a certain kind of applicant. To a comp analyst, the same page is a category with a pay band, built to compare against a thousand other jobs. To a manager, it breaks into small tasks so a person can be scored. To L&D it’s a skills inventory. To the worker, it’s the actual job.
Aderit’s answer is interesting. They don’t force one truth. They keep them all. Comp, talent, L&D, management, and task descriptions live side by side. Each silo keeps its own view. Aderit holds the overlap.
Keep every facet, and something bigger takes shape. Lines run from each version to the job, the worker, the source system, and whoever last changed it. Do it across every record and the lines grow into a web. That web is a knowledge graph for HR. It holds not just the data but how each piece connects to every other piece.
Every field also gets a confidence grade: SOLID, SOFT, SHAKY, or UNKNOWN. That’s data quality that respects the source instead of bullying it into a template. Even an empty field says something, once you figure out why it’s empty.
That’s an answer that thrills a data architect sends an HR generalist running for the parking lot. Aderit’s fix is a filter: show me the job from my seat first, the other views only if I ask.
The job nobody wanted
Under all of this is the least glamorous work in software: integration. Someone once labeled a field address line three and filled it with t-shirt sizes, because it was there. Both founders come out of that world, where budgets die moving data instead of using it.
AI’s best trick is reopening problems people gave up on, and field mapping is the perfect one. It was so dull the industry gave up and let the support queue absorb it. Aderit’s 300-plus connectors and self-healing maps walk back into that abandoned room. Solve it, and a whole layer opens up that nobody could afford to reach before.
First in line are the auditors and whoever writes the next RFP, not the Mobley-versus-Workday crowd asleep at the tiller.
A big crowd will nod politely. A smaller crowd will do them the real favor and try to break the thing. Aderit (aderit.ai) wants the second kind. On one afternoon’s evidence, they can take it.
Correction
The confidence grades — SOLID, SOFT, SHAKY, UNKNOWN — are being built now, iteratively. Same with the full fractal data views such as the job description, comp and talent and L&D and management and task views living side by side. Today we capture the fine-grained detail and evidence, the dynamic contextual views are in flight rather than a maintained model. If we left you with the impression either one was done, that's on us. "Being built" is the accurate phrase for both.