Skip to content
MRJ

Current curiosities

Questions I am in the middle of.

None of these is a finished project. They are the things I am reading about, arguing with myself about, and building small pieces of to find out.

  1. 01

    Can an AI agent operate an enterprise system without becoming an uncontrolled black box?

    Every agent I have shipped so far proposes and never decides. The open part is what happens when the thing it proposes is a write to a payroll, and who has to sign.

    intelligencesystems

  2. 02

    How far can Bangla voice AI run locally, without cloud inference?

    Speech and synthesis already run on a Jetson Orin with a voice cloned from one short clip. The limit I am looking for is the point where quality stops paying for the hardware.

    intelligencemachines

  3. 03

    Can a robot develop useful social interaction from perception and deterministic behaviour alone?

    Zayyan decides when to speak with cooldowns, quiet hours and an interruption budget, not with a model. I want to know how much warmth survives being written as rules.

    machinesintelligence

  4. 04

    What happens when ERP systems become agent-native?

    Forty-two companies currently run on forms. If the form becomes a conversation, the audit trail has to become something else too, and I do not yet know what.

    systemsintelligence

  5. 05

    How should humans and autonomous systems share control?

    A phone agent that falls silent the moment a person joins the call is one answer. It is not a general one, and the general one is what I am after.

    machinesresearch