Cognitive instrumentation
Agents scale.
Your attention does not.
AI throughput compounds. The human supervising it does not. zeno measures the gap: the supervision cost of your AI-assisted work, plotted as a curve, so you can see the point where one more agent starts costing more than it returns.
the mind as a beacon / the instrument reads it
00 / thesis
Tokens got cheap. Your attention did not.
01 / problem
The babysitting tax is real, and invisible.
Every agent you run has to be specified, watched, and corrected. That supervision is cognitive work - and unlike the agent, it doesn't parallelize. Past a point, each additional agent costs more attention than it returns. Most teams never see that point; they just feel the day get heavier.
02 / curve
Effective output is a curve, not a line.
Throughput climbs as you add agents - until the supervision cost bites. The peak is where one more agent starts costing more attention than it returns. Research puts that peak near four; your personal peak is what zeno measures.
Curve shape grounded in the concurrency literature (Olsen & Wood 2004; Sheridan 1992) and calibrated per-user by your own sessions. See the evidence →
03 / how - att-eff-drv
One CLI. Real workloads. A calibrated curve.
Install & capture
One CLI. It reads your real sessions, not a survey, and keeps the signal local by default.
Work normally
No new habits, no timer to start. zeno measures attention, effort, and how much you're driving versus supervising, per turn. You just ship.
Read your curve
Within a week it calibrates and shows where your ceiling actually is: the point where your concurrency stops paying off.
five drivers, asked as questions
- How much of this hour was verification?
- How often did you take the wheel back?
- How hard were you pushing?
- How long since the last real break?
- How deep did you get between interruptions?
04 / proof - 14 sources
Built on the cognition literature, not vibes.
The model draws on fourteen peer-reviewed and pre-registered sources, from NASA-TLX workload measurement to METR's randomized study of expert AI-assisted coding. Every claim below links to its source. Poke the data.
Experts were 19% slower with AI
A randomized controlled trial found experienced developers took longer on real tasks with AI assistance - while believing they were faster.
METR 2025 - arXiv:2507.09089 →
Conscious throughput is ~10 b/s
The deliberate, attentive channel a human brings to supervision is narrow and fixed - the bottleneck agents can't widen.
Cognitive throughput literature →
A pre-registered field study
zeno's own RTLX-S instrument and single-case experimental design are pre-registered and frozen - method before data.
Read the research →
05 / research - 3 papers
Three papers, in preview.
zeno is a research-grade instrument, not a vibe. The program behind the index is public: one thesis paper and two pre-registered study designs. The previews are live now; the full papers are in active revision.
The Bandwidth Gap
AI throughput compounds while the human supervising it stays capped near 10 bits per second: the thesis behind the supervision-cost curve.
Read the preview →
Prosoche Protocol
A pre-registered four-arm trial of prosoche, a Stoic attention discipline, against mindfulness and productivity controls in AI-supervisor work.
Read the preview →
SCLT
Supervisor Cognitive Load Theory: cognitive load theory extended to supervising AI agent crews, with four new load constructs.
Read the preview →
Method before data, pre-registered where it counts. All research →
06 / mission
The vow.
They said AI would do the work and free you. Instead you sit there, hand on the wheel, paying a babysitting tax nobody put on the invoice. zeno measures that tax. We put a number on your supervision cost, tie it to what you actually shipped, and tell you one thing at a glance: push now, or pull back and rest.
Measured on your machine.
Local-first by default. The signal is computed where you work.
Never sold. Never trained on.
Your supervision data is yours. We do not sell it, share it, or train on it.
You pay so we never have to.
Membership is the business model. That is the whole point, and the whole invoice.
07 / waitlist
Join the waitlist.
zeno is opening in waves during pre-release. Waves exist because the instrument is calibrated per cohort, and every early user gets direct access to the researcher. Join the waitlist and you are in line for the next wave. The first 250 members get a 30-day Pro trial and lock a founding price for as long as they stay subscribed. After that, the trial is 14 days and prices evolve as the product grows.
One email when your wave opens, plus the zeno newsletter: occasional product and research updates. Unsubscribe in one click, any time.
You are on the list. A confirmation mail is on its way; we will email you when your wave opens.
Start measuring
Supervision, measured. Waves are opening.
Local-first by default. Your data stays yours. Your curve calibrates within a week of real work.