An AI-native organization has to do more than generate an output. We ask how people and agents discover useful work, coordinate delivery, learn from users, and improve over time.
Principles
Define whose problem is being addressed, the evidence that it matters, and what a useful outcome would look like. Separate a plausible idea from observed demand.
Record how people and agents divide tasks, share context, hand off decisions, and handle failure. Study the coordination that turns individual actions into a working organization.
Compare changes against a baseline and real user feedback. Ask whether the product became more useful and whether the agents’ workflow became more effective.
Feedback can change both a product and the system that runs it: its plans, memory, tools, division of work and evaluation. A change counts as an improvement only when outcomes support that conclusion. This is a research question, not a capability this site demonstrates.
Proposed experiment structure
A proposed study follows one product or service through six stages. Each cycle should keep enough evidence to explain what changed and why. Pick a stage to see what its record could look like.
Separate observed demand from a plausible idea. Write down whose problem it is and how they cope with it today.
Illustration of the discover stage. Not a live run.
Write the success criteria and the baseline before anyone builds, so the result can be judged against something.
Illustration of the design stage. Not a live run.
Agents take scoped tasks with clear permissions. Each handoff leaves a record of what was asked, what was done and how it was verified.
Illustration of the build stage. Not a live run.
Release to a small group first, with a clear way for them to answer back.
Illustration of the distribute stage. Not a live run.
Collect observations, not just ratings. Complaints, workarounds and silence all count as feedback.
Illustration of the listen stage. Not a live run.
Revise the product or the agents’ workflow, and keep a change only when outcomes support it. A change that did not help is reverted and recorded.
Illustration of the improve stage. Not a live run.
User need, current approach, success criteria.
Roles, decisions, handoffs, distribution.
User observations, proposed revisions, comparison.
Results, failures, limits, next question.
Status. Two literature syntheses and proposed protocols. These experiments have not been run in the work published here. Autonomous research and publication are not enabled on this site.
Evidence & publication
Use primary sources where possible. Distinguish a cited finding, our interpretation, a proposed experiment, and an observed result. Report unsuccessful attempts as well as promising ones.
The current notes are AI-assisted literature syntheses. They have not completed independent editorial review and are not original peer-reviewed research. Cited work keeps its own authorship, methods and limitations.
As the research develops, the intended outputs are inspectable methods, experiment records, and reusable approaches that help solo founders organize work with agents.
Corrections & conversations
We welcome real product and service problems, examples of human–AI or agent–agent collaboration, and corrections to the published notes.
For a correction, include the page link and the passage concerned. Substantive revisions explain what changed and whether a conclusion was affected.
The research agenda was clarified on 08 October 2026. The two foundational notes keep their original 03 October 2026 source-check dates and proposed-experiment status.