Writing

WritingPractice recordPublished 16 Nov 2026Yasuhiro Shinsho

Governance Without Unnecessary Machinery

What does an additional check change? A reflection on simplifying the practice’s AI work, distinguishing responsibilities from mandatory stages, and retaining the decisions that matter.

4 min read4 core pointsBilingual
Practice recordDecision materialVerificationHuman-AI workOperating practice

Article

Adding a check may answer an immediate concern. The check can remain, though, after the way the work is done has changed. Alongside doing the work, people or agents accumulate the work of explaining completion to each stage.

When designing how to delegate work to AI, I consider both what may be delegated and which checks need to remain. The number of stages does not establish whether the necessary decisions are being made.

On 15 November 2026, Fragment Practice revised how work is brought to completion when updating its operating rules and code. The revision reconsidered extra coordination and completion stages required for every routine change. Required independent review and checks, actual permission constraints and consequential human decisions remained.

The question was what failure each stage prevented, and what decision or observation it added.

Explaining the same result again

Suppose a change's review and test results already sit alongside its work record. What would be added by requiring those results to be copied into another completion document?

If the audience and the decision are different, explaining the result again has a purpose. Turning implementation checks into material for a business decision is one example.

If the same person would consult the same result to make the same decision, however, the reason for another copy becomes weaker. Having another record does not itself establish stronger assurance.

The useful question is what someone can newly check or decide after reading it.

Separate responsibilities, additional stages

Responsibility for doing the work, checking the change and deciding on publication or business consequences needs to be clear. Whether that always requires an additional named actor or execution stage for each responsibility is a separate question.

For this work in the practice, the revised design allows one capable agent to investigate, implement and perform applicable checks. Required independent review remains separate. The person who holds authority for a consequential publication or business decision still makes it.

What to reconsiderDuplication to reduce, and the purpose to retain
Who carries the work forwardDo not add coordination and completion stages to every routine change. Retain required independent review and decisions by the person with authority.
Explaining completionExisting review and check results need not always be copied into another completion receipt. Findings still need to be resolved and the requested outcome assessed.
Checking the result of an actionDistinguish another explanation from checking the actual state of the target service. The latter establishes a result that the review alone cannot show.

The last check serves a different purpose from another explanatory document. Reviewing a draft or code does not establish what exists at its publication or execution destination. The execution architecture note considers this distinction through task handoffs.

A finding still leaves a choice

Adding another instruction or checking stage after every review finding leaves future work carrying the accumulated responses to past mistakes.

It may be more appropriate to change the code or add a test that catches the same error in later changes. A local correction may be enough in another case. Changing authority requires someone who can make that decision. Where the failure is addressed also changes the maintenance work that follows.

When reconsidering a procedure, I want to examine the failure it prevents, the decision or observation it adds, and how to retain the protection that is needed if the procedure changes. Having been necessary before and remaining necessary now are questions that can be considered separately.

What would show improvement?

The revised instructions establish that stages were reduced. I have not yet measured how that changes human workload or rework. Evaluation also needs to examine whether the necessary checks continue to work when the procedures are used.

Useful observations would include work returning to the founder, corrections after review, operations whose results become uncertain on resumption, and the effort of keeping instructions current. If someone is filling gaps elsewhere after a procedure is shortened, the burden may have moved.

I continue to consider who can reconsider accumulated systems and controls in the practice's own operations. Necessary decisions and their supporting evidence should be easier to find. I want to examine whether reducing the machinery helps, by looking at the work that remains.

Practical entry points

When this theme becomes practical.

If the note feels close to your situation, choose the next entry point below.

Context-specific support

Structure a context-specific issue

Use Services when an active issue needs context-specific structuring around AI governance, security governance, decision material, review points, and responsibility boundaries.

Explore Services

Similar situations

Compare similar situations

Use Cases to compare this theme with common situations where AI adoption, governance, review points, or responsibility boundaries needed structure.

Explore Cases

Related reading

Read next

Related arguments, records and observations to continue with.

Published 26 Oct 2026

Practice Note

7 min read

Who Is Responsible for Reducing Complexity?

Organizations have ways to approve new systems and controls. The authority to reconsider them can be harder to locate. A perspective on operational co…

Practice NoteDecision materialAuthorityReview cycle

Published 9 Jul 2026

Practice Note

8 min read

Design AI Governance as a Review Cycle

AI governance should not be treated as a fixed policy document. As AI products, usage patterns, and organizational conditions change, organizations ne…

Practice NoteAI governanceSecurity governanceReview cycle

Published 16 Nov 2026

Practice record

4 min read

What to Keep When Handing Off Work with AI

What must remain available when an AI conversation ends? Using an article handoff, this note examines how task records, versioned material, live resul…

Practice recordDecision materialVerificationHuman-AI work

Next entry point

From public notes to practical work.

Writing captures the thinking behind decision-ready material. Other notes can be read the same way.