Monthly Newsletter - November 2026
Common Sense Isn’t Always Common Practice – Especially in the AI world
November’s ControlPointAI newsletter connects practical lessons from engineering and contract oversight to AI governance, explains the evolution of our mission and vision, and reports progress on our piping UT and EU AI Act hiring demonstrations. You can delegate the analysis. You cannot delegate the accountability.
By Wayne Couch ·
Read the damn contract.
That sounds obvious. So do the other basics: make sure your offering meets the requirements, meet the deadlines, verify what AI tells you, and read the original source yourself.
Yet these are often the steps people skip.
As a government representative, I reviewed a lot of proposals and bids over the years. My own rule of thumb: do these few things consistently, and you’re already ahead of half your competitors. You’d be surprised how many bids fall short on the basics.
AI makes it easier to produce an answer, a proposal, or an analysis quickly. The person responsible still has to determine whether the work is correct and meets the actual requirements.
A convincing summary can overlook a qualification. A polished proposal can miss a deliverable. A confident answer can be wrong.
These principles started as a plain-language reminder of fundamentals I have relied on in engineering, contract execution, and technical oversight throughout my 37 years of government/DOD service. Working with AI has given me more reasons to keep them visible.
The Principles
- Read the actual contract, including its specifications, terms, conditions, and deliverables.
- Make sure your offering meets what the customer is actually buying.
- Meet every specified deadline.
- Read the contract yourself. Use AI to assist your review, while retaining responsibility for understanding what the contract actually requires.
For working with AI
- Verify what AI tells you. Confidence is not evidence of accuracy.
- Read the original sources. An AI interpretation should help you examine the evidence, rather than become your only connection to it.
The underlying point is straightforward:
You can delegate the analysis. You cannot delegate the accountability.
From Mission and Vision 2.0 to 3.0
In Newsletter 3, “Technology Is Moving Faster Than Authority,” published February 27, we introduced Mission and Vision 2.0 under our earlier name, Human-Centered AI Integration. It established our focus on authority design: explicit delegation, enforced boundaries, and accountable human decision-making in high-consequence environments.
Mission and Vision 3.0, issued October 4, carries that direction forward and makes our operational method more explicit.
The mission now describes how we work: map the process with the people who operate it, trace how information moves, identify where decisions occur, and establish who has authority and where independent review is required. From that understanding, we identify gaps in authority, evidence, and accountability and define control points for review, hold, escalation, stop, and reauthorization.
The vision also broadens an important question. Version 2.0 stated, “If AI makes a decision, we should always know who is accountable.” Version 3.0 asks us to examine wherever AI influences an outcome, tracing the evidence, responsible human authority, and controls governing the work.
That distinction matters. A human may formally approve an action while relying on incomplete evidence, an incorrect AI summary, or a recommendation they have not adequately examined. Human oversight therefore depends on that person’s qualifications, information, authority, and practical ability to intervene.
The commitment to accountable human decision-making remains. Version 3.0 makes the work needed to support it more visible.
That is part of what we mean by Beyond the code.
Demonstration Project Updates
Our two demonstration projects explore how process mapping and human oversight work in different operational settings.
Piping UT Demonstration Project
The published engineering drawing gives readers a concrete view of the information flows and control points in our synthetic piping ultrasonic testing case.
The demonstration examines how evidence, qualified human review, and decision authority connect as work moves through the process. It also addresses ambiguity: when an outcome cannot be established, the process needs to preserve that uncertainty and provide a path for review.
The next implementation focus is the first mapped handoff, Node 1 to Node 2. The work ahead is to turn those requirements into a working control and test whether it behaves as intended.
View the Piping UT Demonstration engineering drawing.
EU AI Act Hiring Demonstration
The hiring demonstration applies the same operational discipline to an AI-assisted employment process, where recommendations can affect a person’s opportunity for work. Its scope draws on Annex III, point 4(a), of the EU AI Act, which identifies recruitment and selection uses, including filtering applications and evaluating candidates.
During October, we established the project’s legal ground rules, developed the functional sequence, and identified an initial set of requirements to support the first process drawing. That drawing is the next major deliverable before publication on the website.
A central focus is human oversight under Article 14. Paragraph 4 addresses understanding the system’s capabilities and limitations, recognising automation bias, interpreting outputs, and being able to disregard, override, reverse, or stop the system’s operation where appropriate.
Article 26(2) adds the organisational responsibility: assigning oversight to people with the necessary competence, training, authority, and support. Our demonstration examines what those responsibilities require in an actual hiring workflow.
We are also examining how an AI recommendation can influence an outcome even when a human formally makes the decision—a concern explicitly addressed through the automation-bias provision in Article 14(4)(b). A recorded approval alone does not establish that meaningful review occurred.
The work is translating these requirements into review responsibilities, control points, and evidence requirements that can be examined as the demonstration develops.
Looking Toward 2027
We are also developing the ControlPointAI strategic plan for 2027.
That means deciding where to concentrate our effort, what the demonstrations need to establish, and how the resulting work can become useful to others. We may also develop a continuing plain-language series on these fundamentals, using practical examples from contracts, engineering, and working with AI.
The decision on whether to form an LLC has been postponed. We will revisit business formation as the strategic direction develops.
The same discipline applies to our own planning: understand the requirements, examine the evidence, and make a deliberate commitment.
There is plenty of discussion about what AI might do. Our contribution is to work through what responsible use requires in an actual process—and make that work visible enough to examine.
Beyond the code.
Wayne Couch ControlPointAI