AI compounds. Human attention does not.
Zeno measures the supervision cost of your AI-assisted work. Install the CLI, wrap your sessions, watch your babysitting-tax curve calibrate within a week. Decide what's worth your attention.
The dashboard
Identity flows over your existing tailnet. The dashboard reads your sessions and renders the babysitting-tax curve from your real work. No telemetry leaves your network.
// app.zeno.center · tailnet-only during early access · watch the guided tour →
How it works
Three steps, ten minutes to first signal, one week to a calibrated curve. Pre-launch: CLI install lands on PyPI for early-access invitees.
-
Install the CLI
One command. Python ≥ 3.12. Sends to your private tailnet endpoint, never the public internet.
pipx install zeno-cli
-
Wrap your sessions
Mark when you start supervising agents and when you stop. Works with any harness: Cursor, Claude Code, Continue, Cody, custom.
zeno session start --project myrepo # ... run your agents ... zeno session end
-
Read your curve
After five to ten sessions the dashboard renders your personal babysitting-tax curve. See where N-agents-active inflects and decide what to cut.
open https://app.zeno.center
Why we built this
AI capability scales faster than human deliberate throughput. Three measurements anchor the gap:
-
In expert repository tasks, AI assistance measured as a 19% slowdown despite perceived speedup.
-
Deliberate human throughput is around 10 bits per second.
-
Knowledge workers face roughly 275 interruptions per day in modern work telemetry.
Get early access
Pre-launch. CLI invites go out in waves; research updates ship weekly. Pick what you want.
Solo Pro for individuals. Team plans in design - email admin@zeno.center if you have a multi-engineer use case.
Research foundation
Zeno is built on a research thesis with three published dossiers. The CLI and dashboard operationalize what these document.
-
The Bandwidth Gap outlines the operating thesis: AI capability scales faster than human deliberate throughput. Cognitive debt, babysitting tax, and cognitive sovereignty.
-
Prosoche Protocol defines a pre-registered RCT pathway for testing attention interventions in AI-saturated work.
-
Supervisor Cognitive Load Theory (SCLT) extends CLT to multi-agent supervision with explicit load categories.
Questions people ask us
What does Zeno actually measure?
Sessions of supervised AI work. The CLI records how many agents you were running, your composite cognitive load while supervising them, and the output quality you saw. The dashboard renders your babysitting-tax curve and weekly aggregates.
Does it work with Cursor, Claude Code, Continue, or Cody?
Zeno measures your supervision pattern, not the tools themselves. Wrap any AI-assisted session with the CLI, whatever the harness. The data model is harness-agnostic.
What data leaves my machine?
Session metadata: number of agents active, composite load probe responses, output quality ratings. No code, no prompts, no model outputs. The endpoint is private to your tailnet during early access.
How is this different from time tracking?
Time tracking measures hours. Zeno measures supervision load and output quality, the two variables that decide whether AI is a multiplier or a tax. You can't get there from a stopwatch.
Why is the dashboard tailnet-only right now?
Privacy by default. Early access ships the API and dashboard on your tailnet so identity and traffic stay on your network. Public sign-in lands when the proxy-node graduation ships, planned for the first non-Tailscale paying user.
Contact
Business contact: admin@zeno.center.
Zeno Center B.V. · Eindhoven, Netherlands.