I help companies see how they actually run, and decide where AI belongs.
The knowledge of how a company works is not in the org chart, the process document, or the system logs. It is in the people doing the work — and nobody asks all of them, properly, and keeps every word. I built ProcessOre to do exactly that: the part no human can. The plan on top of it — what to stop, what to fix, what to automate, and what it returns — is the part I do myself. I have spent my career in operations and product, and I build automations and put AI into companies.
Now
ProcessOre
Operational discovery: a system that interviews everyone in a business function — every person, every follow-up, every word kept — and builds the map of how the work actually happens from what they said. On top of it, the plan: what to stop doing, what to fix, and where AI will pay for itself. The system does the part no human can. The plan is the part I do. Taking on the first design partners now.
Bercilak
A memory app you talk to. You send it a thought — typed or spoken, from wherever you already are — and it keeps it somewhere you can actually find it again.
Positions
Seven things I will defend, and argue with anyone about.
AI should do what no human can. People should decide what no model can.
No person can interview forty people, ask every follow-up, and keep every sentence — that is the system’s job. No model can say what will actually work in a company, because that takes experience of what survived, and a model has memory. The mistake is letting either one do the other’s job.
Most AI programmes stall long before the model is the problem.
The blocker is almost never capability. It is that nobody in the building can say which work is worth pointing it at — and the people who could say are three levels down and were never asked.
A description of how work happens is worthless without the person who gave it.
Every statement about a process came from someone, on a day, in a sentence. Systems that record only the conclusion cannot be repaired when they go stale — they can only be redone, which is why they never are.
Deleting work beats automating it, and it is the step everybody skips.
Automating a step that should not exist makes the wrong thing permanent, faster, and much harder to argue with afterwards. The order is delete, then fix, then automate — and the first one is free.
A recommendation to delete something should be a hypothesis, not a verdict.
Say what you found, name who said it, and name the check. The approval nobody can recall ever rejecting anything is sometimes a control that somebody gets audited on, and finding that out afterwards is the expensive way.
The people doing the work already know where it breaks. They are just never asked properly.
They are asked in a workshop, in a room, in front of their manager, where the honest answer is a career risk and the safe answer is the one on the org chart. Ask them privately, in writing, in their own time, and you get a different set of facts.
“We just handle it case by case” is a finding, not a gap.
A map that says nobody could establish who owns a step is worth more than one that guesses. The guess reads as completeness and is the thing that quietly makes the whole document untrustworthy.
Contact
If you want to know how your company actually runs before you put AI on it, or you want to be one of the first companies ProcessOre runs on, write to me. Investors too.
mj@marijusjankauskas.com