The research notebook / an authored demonstration

The unfinished.

A good essay leaves something unresolved. This is a place to stay with the question a little longer.

Three roles, one small inquiry. A deliberately authored simulation: no live agents, external research, credentials, or background jobs. Nothing here is a new finding.

01 / Questions still open

What would change
our minds?

Questions to investigate, not a queue of work being performed.

Browse the essays ↗

Capability
What happens after the impressive first demo?

Can a two-person AI-assisted team maintain a product when the work shifts from making to repairing, supporting, and deciding?
Judgment
When output gets cheaper, where does expertise show up?

What would distinguish a well-chosen problem from a well-produced answer to the wrong one?
Accountability
Whose work disappears from the picture?

How should a claim about small-team capability count platform labor, borrowed infrastructure, and people outside the team?

02 / A worked example

One question, three ways of looking.

Starting position: Qubit Teams: The Future Built by Two. An essay is an argument to examine, not empirical evidence.

I / Possibility

The scout

Find the claim worth testing. Separate what an essay proposes from what it can establish.

II / Counterpoint

The skeptic

Look for the missing denominator. Ask what the attractive explanation leaves outside the frame.

III / Next question

The synthesist

Keep the uncertainty. Turn the disagreement into a question that evidence could actually answer.

Milestones / authored sequence

Frame a testable claim Not started
Expose an assumption Not started
Leave an investigation brief Not started

Observe the reasoning, not a live feed

The complete authored sample is readable below without JavaScript. Interactive controls, when available, reveal three fixed stages.

Three predetermined stages, about two seconds each. No network requests. Step through at your own pace instead.

The demonstration has not started.

No observations yet. No evidence has been collected.

01 / Scout / authored note

Make the claim small enough to test.

The essay proposes that AI could expand a very small team’s capabilities. For this example, narrow the inquiry to one product’s maintenance after launch—not every kind of work.
02 / Skeptic / authored note

Count the work beyond the demo.

A fast launch would not by itself establish durable capability. Support, incidents, vendor services, and help from other people could change the story. These are possible confounders, not observed failures of a particular team.
03 / Synthesist / authored note

Trade a sweeping claim for a better question.

Ask what two people can sustain over time, under what conditions, and with whose help. The resulting sample brief names what to look for, without pretending the study has happened.

03 / Resulting sample artifact

Authored simulation
Unreviewed study proposal

No participants, measurements, external evidence, or research results are represented here.

Beyond the first impressive demo.

Explicit question Under what conditions can a two-person AI-assisted team sustain a product after launch without shifting uncounted work to others?

Proposed next investigation: Follow consenting teams from launch through a defined maintenance period. Document product scope, prior experience, AI use, external help, and responsibilities before comparing outcomes.
Potential evidence needed: Dated maintenance and incident records; support and contractor hours; changes in product scope; interviews with the people doing the work; and a comparison group with similar responsibilities. Include stalled or abandoned projects, not just successes.
Limitation: This brief is authored in advance from a single essay’s premise. No sources have been gathered and no teams studied. Even a future comparison could be confounded by experience, product complexity, selection bias, or unrecorded help. It would not alone prove that AI caused any difference.
What could change the question?: If maintenance depends mainly on scope and outside support rather than headcount, study those conditions instead of treating “two people” as the explanation.

Original starting essay: Qubit Teams: The Future Built by Two ↗

04 / Method & boundaries

What this is.
What it is not.

Every role, note, milestone, and sample paragraph is authored and committed with the site. Start reveals those same notes in order; Pause holds your place, Resume continues, and Reset returns to an empty notebook. Switching away pauses playback until you resume. Reduced-motion preferences use one step at a time.

The cycle does not run models, search the web, collect credentials, or call external services. Its state lasts only for this page visit. The role illustrations are original static line studies, not data visualizations or indicators of live activity.

Public evidence register: empty.

No public-data records have been ingested. A committed boundary in content/research-example.json describes a possible manually reviewed record: source URL, source title, retrieval timestamp, and a plain-text excerpt. It is documentation, not an active importer. A source’s provenance would not establish its reliability.

Any future source record would need validation and editorial review before publication. No arbitrary HTML, hidden network access, or automated research jobs are part of this demonstration.

What would change our minds?

What happens after the impressive first demo?

When output gets cheaper, where does expertise show up?

Whose work disappears from the picture?

The scout

The skeptic

The synthesist

Make the claim small enough to test.

Count the work beyond the demo.

Trade a sweeping claim for a better question.

Beyond the first impressive demo.

What this is. What it is not.

Public evidence register: empty.

What would change
our minds?

What this is.
What it is not.