The Human and AI Company
How to organize around the customer, decide who owns AI, and move people from doing the work to directing it. A practical executive paper by Aaron Dillingham.
Executive paper · 17 pages · October 2026 · Aaron Dillingham
From AI tasks to customer outcomes
Redesign the work.
Then decide where AI belongs.
A faster draft does little for a customer if the request still waits in the same approval queue. This paper examines how to organize the whole flow: the customer job, the people responsible, the agents involved, and the controls behind each decision.
Written for chief executives and operating leaders of midsize companies, it connects organizational design with a practical way to pilot and evaluate AI.
Open the PDFAI assists
A person does the job, with AI supporting individual tasks.
Human in the loop
AI prepares work; a person approves each result before release.
Human on the loop
Agents handle bounded routine work; people direct, monitor, and handle exceptions.
More autonomy is a permission to earn through evidence, not a starting assumption.
Inside the paper
Eight steps.
One operating design.
Follow the approach from the first customer request through ownership, team design, controls, and a measured pilot.
01Find the job worth changing
Choose recurring work with a clear customer result. Walk real cases, identify the delay, and establish the baseline before choosing a tool.
02Give AI a defined role
Map the full sequence, compare a simpler non-AI option, and give each agent explicit inputs, outputs, permissions, and limits.
03Organize around the job
Name one owner for the customer result. Define how specialists, department managers, technical owners, and agents work across handoffs.
04Decide who owns AI
Set common controls while allowing useful building close to the work. Each agent needs a named owner and a clear route for resolving decisions.
05Bring your people with you
Agree workload, learning, incentives, feedback, and responsibilities before launch. Protect the team's ability to operate and judge the work.
06Set controls for the real job
Define source information, release checks, review capacity, support, and a tested manual fallback. Test whether the controls catch meaningful errors.
07Let AI earn its autonomy
Separate decisions by consequence. Extend permissions only after measured results, with recorded overrides and a way to withdraw trust.
08Prove it, then scale
Compare the current process, a simpler improvement, and the AI design. Count review, upkeep, queues, and rework before deciding what to expand.
Included in the PDF
A worked example.
A one-page operating brief.
The illustrative proposal example follows a workflow from redesign to supervised operation. The operating-brief checklist brings together the customer job, assignments, owners, team, information, controls, permissions, pilot measures, and costs.
The example uses invented figures to explain the approach; it is not a consulting client result.
Download the 17-page PDFThe separate editable workbook and application prompt pack remain in development.
Evidence & limits
A framework to test in your business.
The paper includes its research sources and explains their limits. Those sources inform the design choices; they do not establish the method's effectiveness in a particular company or guarantee a return. A pilot should establish the evidence for the actual workflow.
Apply the approach to a workflowPeople. Systems. Technology.
What keeps slowing
your business down?
Bring the problem or the project. We'll discuss how AI, automation, and clearer responsibilities could help your business.
Discuss your project