private twin
Language FR EN

The project

Where it comes from, and how it is built

Private Twin did not start with development. It started with a problem lived as a manager.

The name

Janus faced two directions

Engraving of Janus after a Roman coin: two bearded faces turned back to back, inside a circular medallion
Janus, after a Roman coin
Private Twin mark: two profiles, one solid, one outlined
The Private Twin mark

In Roman mythology, Janus is the god of beginnings, transitions, and duality — looking simultaneously to the past and the future. The mark carries the same idea, literally: two heads. The solid one is you. The outlined one is the twin: a private reflection, faithful but distinct, that remembers and learns. One twin per person, and it stays private.

The product does what the name says: it looks at what was planned and what actually happened, at the same time. The space between the two is where the record keeps itself current.

Like Janus, Private Twin guards the threshold of time: yesterday’s lessons and tomorrow’s goals.

Method

Six steps, from research to the field

The project did not start with development. It started with a problem lived first-hand.

  1. 01

    Explore the problem

    Research

    Studying research on agentic AI and cognitive science, in particular human attention and the effects of fragmented work.

  2. 02

    Frame the idea

    Lived problem

    Running several projects involving many people, while information, decisions and actions sit scattered. In that setting, reliable and current visibility becomes critical to coordinate and to anticipate risk.

  3. 03

    Test initial interest

    Field

    Calls and video conversations with about ten people, to test the problem against their day-to-day and check the subject held.

  4. 04

    Build

    Over a year

    Building a complete prototype and an AI-assisted development method, plus a harness, the technical frame that runs, controls and evaluates the agents. It keeps language model calls to what is strictly needed and avoids depending on a single provider.

  5. 05

    Monitor and evaluate

    Build

    Building Sim Lab to keep a complete trace of runs, follow how responsibility is delegated between human and agents, replay scenarios and catch regressions.

  6. 06

    Deepen market validation In progress

    Field

    Currently under way with four families of knowledge workers: twenty interviews in total, comparing how sharply the problem is felt, how acceptable capture is, and what follow-up actually happens.

Sim Lab

Checking that an agent behaves as intended

An agent that convinces over a few runs is not necessarily reliable over time. Sim Lab freezes a working day, replays it and compares versions.

As on a production line, each step is a control point. When a result degrades, Sim Lab helps identify where and why the behaviour changed.

  1. Snapshot A day and the state of the system, frozen.
  2. Duplication The original World stays intact.
  3. Replay Same scenarios, same conditions.
  4. Comparison Code, configuration, prompts or models.
  5. Divergence The first point where behaviour degrades.

Who builds it

A project built by Garry Jakubiak

Private Twin is built by Garry Jakubiak, GenAI transformation consultant, from a problem he ran into himself as a manager. One person carries the project today.

Background and practice on garryjakubiak.com