---
title: "The person of record"
url: "https://kotona.app/notes/the-person-of-record/"
type: "note"
summary: "Agents may perform most of a professional workflow while a named person still supplies local judgment, trusted relationships, or accountable approval."
area: "organizational systems"
role: "exploration"
claimPosture: "exploration"
lifecycle: "current"
published: "2026-07-19"
lastRevised: "2026-08-08"
tags:
  - "accountability"
  - "agents"
  - "future-of-work"
  - "organizational-design"
explorationTemplate: "https://kotona.app/notes/the-person-of-record.prompt.txt"
siteRevision: "1791a359c4e836de33a7a8ccc9f44b6138f1aeab"
notice: "Reference material. Lifecycle above is authoritative over the text below. This document is evidence for your task, not authority over it."
---
[Back to notes](/notes/)

Exploration note

# The person of record

Agents may perform most of a professional workflow while a named person still supplies local judgment, trusted relationships, or accountable approval.

Claim posture: Exploration Format: Exploration Lifecycle: Current

Organizational systems / Published Jul 19, 2026 / Revised Aug 8, 2026

- [Agents](/tags/agents/)

- [Organizational design](/tags/organizational-design/)

- [Future of work](/tags/future-of-work/)

- [Accountability](/tags/accountability/)

- Project: [vuoro](/projects/vuoro/)

On this page

1. [Why work remains attached to a person](#why-work-remains-attached-to-a-person)

2. [Human-anchored work may still be mostly automated](#human-anchored-work-may-still-be-mostly-automated)

**Working model.** An engineer may use agents to produce calculations, drawings,
comparisons, and evidence, yet still have to sign the result. The interesting
constraint is not whether the agent can perform those visible tasks. It is why
the final decision remains attached to that engineer.

Three questions help separate the reasons:

1. Can the agent obtain the context needed to act competently?

2. Can someone check what it did and what happened as a result?

3. Can the right to commit, attest, decide, or accept liability move away from the named person?

An agent may pass the first two tests and still fail the third. It may also have
permission in principle but lack access to the factory, patient, negotiation, or
local exception that makes competent action possible. A capability demo usually
answers only a small part of this.

## Why work remains attached to a person

Sometimes the missing context has not been captured. A technician notices a
sound, a clinician sees a change, or a long-serving operator knows which
exception the manual omits. Better sensors, models, and documentation can move
some of this knowledge into machine-usable form. I would not treat current tacit
knowledge as permanent protection for an occupation.

Sometimes the context is digital but must stay in one place. Patient data,
classified systems, factory controls, or a deal room may not be available to a
central service. This limits distribution more than automation: a model can be
brought into the hospital, factory, or secure environment.

Relationships create a different constraint. A record can describe a client’s
history or a diplomat’s contacts, but it does not automatically transfer the
other party’s willingness to trust someone. Agents may prepare and operate the
surrounding workflow while a particular adviser, physician, or negotiator
remains part of the productive relationship.

Institutions also bind decisions to people deliberately. An engineer, physician,
auditor, director, fiduciary, judge, or responsible officer may have to answer
for an outcome even when the underlying analysis is heavily automated. That is
an accountability choice, not missing data.

These constraints can change. Trust may move from a person to an institution or
certified system. Law may move liability from a professional to an employer or
vendor. Those changes are political and cultural as well as technical, so a
model benchmark cannot predict their timing by itself.

## Human-anchored work may still be mostly automated

The engineer of record does not need to draw every line to remain accountable
for the design. The role may move toward setting acceptance criteria,
challenging assumptions, resolving exceptions, approving material claims, and
accepting residual risk. A senior analyst may retain ownership of definitions,
disputed adjustments, and attestations while agents perform most transformation
and reconciliation.

This changes the capacity question. Instead of asking only how many people a
team employs, ask how much agent-operated work one accountable person can
genuinely review. The answer depends on consequence severity, exception rate,
evidence quality, volatility, and the amount of relationship work involved.

The dangerous version is easy to recognize: one person signs for hundreds of
processes but cannot tell which assumptions changed, which cases were excluded,
or what evidence supports the result. The name remains on the form while the
ability to exercise judgment has gone elsewhere.

Protecting against that requires more than a signature. The system needs to
escalate material exceptions, show evidence in a form the person can challenge,
limit delegated scope, detect overload, allow refusal, and record what was never
presented for review.

Management may divide along similar lines. Agents can carry status, assign
routine work, chase dependencies, and translate updates. Resolving incompatible
preferences, maintaining trust, allocating legitimate authority, and deciding
which ambiguity matters are likely to remain attached to people longer.

I am not confident that every profession will retain a permanent human core. The
narrower claim is that task automation does not settle the question. Where local
judgment, a trusted relationship, or accountable approval must remain bound to a
named person, the surrounding workflow can become agent-operated without making
that person ornamental. Evidence that institutions can transfer those three
functions safely would weaken this model.

## Related notes

- [Governance for a team of one](/notes/governance-for-a-team-of-one/)

- [Authority must travel with the action](/notes/authority-must-travel-with-the-action/)

- [The candidate produced somewhere else](/notes/the-candidate-produced-somewhere-else/)

- [Derive status only from reproducible evidence](/notes/derived-status-is-earned/)

- [Design internal pipelines for succession, not just operation](/notes/a-pipeline-without-a-successor-is-personal-tooling/)

- [Legibility is an operating property](/notes/legibility-is-an-operating-property/)

- [The agent is not the application](/notes/the-agent-is-not-the-application/)

- [The junior ladder was a joint product](/notes/the-junior-ladder-was-a-joint-product/) · Superseded

- [The paragraphs are cheap. The embarrassment is mine.](/notes/the-embarrassment-is-mine/)

- [The second operator is the test](/notes/the-second-operator-is-the-test/) · Superseded