05 / Action classes

Persistent agents need persistent boundaries.

Frontier agents can research, browse, use tools and continue work without a fresh prompt. More useful than a chat window—and capable of carrying mistakes further.

I treat an agent like a worker with a computer: give it a job, a bounded identity, minimum tools, memory appropriate to the role and explicit authority for what it may change.

Agents may read and propose broadly. Authority should remain narrow.

Memory is not policy. A bot is not the system of record.

An independent operating-model exercise. Example roles and workflows are proposed designs, not an employer implementation.

The worked case ↓PermissionsRisks

01 / Memory gap

Information exists. Organisational memory does not.

An Account Manager learns that availability changed, documents are missing, a payout is disputed or a promise was made. It might arrive by email, phone, video, in person, hypothetical WhatsApp or platform message.

The note stays on paper, in personal email, or in a CRM another team does not check. A colleague acts without it. The customer repeats themselves, receives a conflicting answer and opens a ticket. The Account Manager is pulled back in.

Communication to organisational memory A permitted communication event is linked to an account. An agent extracts sourced observations. A named owner or explicit policy reviews the proposed change before it enters the governed record. Permitted business event Known account + relevant context Extract with sources Observation · uncertainty · object ID Review candidate change Named owner + permitted action Governed record Current state + correction + audit Communication to organisational memory A permitted communication event is linked to an account. An agent extracts sourced observations. A named owner or explicit policy reviews the proposed change before it enters the governed record. Permitted business event Known account + relevant context Extract with sources Observation · uncertainty · object ID Review candidate change Named owner + permitted action Governed record Current state + correction + audit
The agent does not become the memory. It helps move useful information into the memory the organisation already governs.

This is the same missing-owner problem explored in the knowledge and tooling case. The goal is not to read everyone’s mailbox. It is to move useful facts into governed organisational state.

02 / Capture

Capture the event, not everyone’s inbox.

Avoid universal hidden BCC into a shared AI mailbox.

It collects too much: personal or confidential conversations, duplicate attachments, secrets and casual speculation. Retention, correction, deletion and employee-monitoring concerns become harder to resolve—not easier.

Prefer native integration

A business-system connector receives permitted email, calendar or platform-message events and resolves them to known accounts. Extract bounded fields; do not permanently mirror whole mailboxes.

Make capture explicit

“Log to Account”, a dedicated company workflow or structured post-call notes. Meeting capture follows attendee consent and applicable organisational policy.

Journaling is conditional

Only for a justified audit context, after legal and privacy review, including works-council review where relevant. Define retention, access, personal-message exclusion, redaction and audit first.

03 / Memory design

Two kinds of memory.

Agent memory and organisational memory Two conceptual authority boundaries, not a requirement for two databases. Working memory helps an agent perform a job. Governed organisational records define current truth. Agent memory Role preferences · task history Working hints · temporary context Organisational memory Current policies · approved knowledge Account / Listing / Booking state Reopen the current source Process hints help retrieval. They cannot override truth. Agent memory and organisational memory Two conceptual authority boundaries, not a requirement for two databases. Working memory helps an agent perform a job. Governed organisational records define current truth. Agent memory Role preferences · task history Working hints · temporary context Organisational memory Current policies · approved knowledge Account / Listing / Booking state Reopen the current source Process hints help retrieval. They cannot override truth.
An agent may remember process. The organisation owns truth.

Agent memory answers “How do I usually perform this job?” Organisational memory answers “What is true right now?” Approved pricing, compliance rules and audit records belong with their accountable source owners.

The bot may remember how to find policy. It may not become policy.

When less memory is better

A narrow research job can use a request, public sources and a temporary workspace: research → cited report → expiry. No internal access or writes. An ephemeral design has less state to govern. Risk determines persistence; dual memory is not universally superior.

04 / Permissions

Permission is a product decision.

A bounded permission stack Each agent has a purpose, owner and identity. Its retrieval, tools, network and actions are scoped. Approval checks, execution budgets and retention limits constrain the job. Identity + accountable owner One purpose; explicit data scope Tools + network scope Minimum access, not wildcard access Actions + approval rules Check authority at the tool boundary Budgets + retention Time / cost limits · logs · stop / revoke A bounded permission stack Each agent has a purpose, owner and identity. Its retrieval, tools, network and actions are scoped. Approval checks, execution budgets and retention limits constrain the job. Identity + accountable owner One purpose; explicit data scope Tools + network scope Minimum access, not wildcard access Actions + approval rules Check authority at the tool boundary Budgets + retention Time / cost limits · logs · stop / revoke
A role description is not an access-control boundary.

No general company bot with inherited access to everything. Each job needs its own identity boundary, owner, purpose, data scope, tools, network scope, permitted actions, approval rules, cost/time budget and retention policy.

Illustrative roles and permissions
Team Useful work Read access Write authority
Marketing Research / campaign drafts Public web + approved analytics Draft only
Sales Account research / meeting brief CRM account scope Draft / proposed task
Account Management Portfolio brief / follow-up / quality flags Assigned accounts + approved mail context Notes/tasks behind policy gate
Booking / Care Case summary / retrieval / reply draft Ticket + Booking No money or booking commit
Trust & Safety Evidence packet / contradictions Bounded risk / identity view No final KYC or ban
Finance Reconciliation explanation Bounded ledger read No payout or refund
Platform Ops Health / workflow analysis Approved operational read Controlled admin gate
Leadership Executive synthesis Governed aggregates None

05 / Action classes

Read broadly only within the job. Commit narrowly.

Read, propose and commit Reading and drafting stay within a job scope. A separate authority check determines whether an action may execute. Missing authority means no mutation; permitted execution produces an audit receipt. Observe → draft Approved sources → reviewable proposal Authority check Identity + scope + state + evidence Not authorized No change; return the blocker. Authorized Bounded commit → receipt Read, propose and commit Reading and drafting stay within a job scope. A separate authority check determines whether an action may execute. Missing authority means no mutation; permitted execution produces an audit receipt. Observe → draft Approved sources → reviewable proposal Authority check Identity + scope + state + evidence Not authorized No change; return the blocker. Authorized Bounded commit → receipt
Autonomy should increase only where the blast radius is understood.
Four action classes
Class Examples Authority
0 · Observe Public search, approved article, scoped analytics, case summary No per-action approval inside the approved read scope
1 · Draft Reply, proposed CRM note, report, routing suggestion Review queue; no external send
2 · Bounded reversible Internal task, safe label, proposed note, ticket routing Explicit policy, audit and rollback only
3 · Sensitive Payout/refund, booking, KYC, ban, contract, deployment, public policy Deterministic checks and explicit human authority; no autonomous promotion by default

06 / Five jobs

Start with five jobs.

Research

Public web and approved sources → cited packet. No send or internal mutation.

Draft

Replies, internal notes and knowledge candidates. Never sends directly by default.

Account brief

Pre/post-meeting packet. Reads only assigned accounts and approved context.

Ops brief

Scheduled queue changes, SLA risks, knowledge drift and recurring failures.

Coordinator

Routes work and combines one review packet. Coordination rights are not the union of every worker’s permissions.

07 / Worked scenario

Capture the promise before it becomes a ticket.

Hypothetical email: a landlord says Property X will be unavailable next month and payout details are changing. If that stays in an inbox, a live listing can become a booking conflict—and “I told you” becomes another investigation.

Fact, inference and proposed action A landlord statement is an observation, not verified platform truth. The possible conflict is an inference. A proposed availability update needs review by the assigned account manager. Payout details require a separate verified finance process. Observation “The unit is unavailable next month.” Source + author + date + Listing ID Inference A future booking may conflict. Check dates and current state. Proposed action Prepare an availability update. Required owner: assigned AM. Separate sensitive request Payout-detail change → verified finance workflow. No account detail changed from email alone. Fact, inference and proposed action A landlord statement is an observation, not verified platform truth. The possible conflict is an inference. A proposed availability update needs review by the assigned account manager. Payout details require a separate verified finance process. Observation “The unit is unavailable next month.” Source + author + date + Listing ID Inference A future booking may conflict. Check dates and current state. Proposed action Prepare an availability update. Required owner: assigned AM. Separate sensitive request Payout-detail change → verified finance workflow. No account detail changed from email alone.
Extraction is not authority.

The designed path: permitted email event → source-linked facts → resolve Account / Listing / Booking → compare current state → owner preview → permitted update → receipt and follow-up.

The value is not “reading email”. It is recognising when a conversation changes an organisational object, then making that change reviewable.

08 / Knowledge drift

Knowledge candidates, not knowledge autopilot.

Detect a knowledge conflict without publishing it The agent compares approved knowledge and scoped case evidence, produces a source-linked discrepancy report and recommends an accountable owner. It cannot overwrite public policy. Compare approved sources Public article · internal template Ticket themes · correction notes Conflict packet Sources + affected scope + candidate answer Canonical owner + proposed review Owner reviews policy Approve correction or reject proposal. No customer-facing change before approval. Detect a knowledge conflict without publishing it The agent compares approved knowledge and scoped case evidence, produces a source-linked discrepancy report and recommends an accountable owner. It cannot overwrite public policy. Compare approved sources Public article · internal template Ticket themes · correction notes Conflict packet Sources + affected scope + candidate answer Canonical owner + proposed review Owner reviews policy Approve correction or reject proposal. No customer-facing change before approval.
A discrepancy is a reason to investigate, not permission to publish.

Illustrative conflict—not a real policy: an article says 48 hours, a template says 72, and recent replies say 24. The output is “conflict found”, sources A/B/C, affected accounts, Customer Operations as proposed owner, and a request to review the article and template. No customer-facing change made.

Five Account Managers explaining one rule differently is an operating signal. The candidate canonical answer still needs the actual policy owner.

09 / Routines

Useful work before someone remembers to ask.

Morning operations

Queues, SLA breaches, account promises, stale knowledge and incidents → one packet ordered by risk and owner.

Before the meeting

A scheduled brief: account, conversations, bookings, tickets, payout exceptions and commitments. Only permitted context.

After the meeting

Approved notes → observations → commitments → proposed CRM updates and tasks → owner review.

Weekly knowledge check

Repeated questions and contradictory answers → source-linked knowledge candidates.

Where this helps

Context assembly, repeated routines, parallel research, richer handoffs and less copying between tools.

After-hours preparation

Prepare a useful queue without sending messages or mutating sensitive state.

10 / Risks

Persistent access means persistent risk.

Credential inheritance

A computer-use worker can act through real logins. Use a dedicated work identity and minimum access.

Prompt injection

Treat external content as data, not runtime instructions. Enforce tool permissions outside the model.

Memory poisoning

Stale or malicious context persists. Reopen canonical sources; allow review, correction and expiry.

Cross-context leakage

Retrieve only information relevant to the role. Separate identities where isolation is required.

Shadow actions

A draft worker gains send/write tools. Deny those capabilities at the tool boundary.

Retention

Keep context only for a defined purpose and period, with correction and deletion handling.

Runaway execution

Bound time, tokens, cost and tool calls. Stop repeated failures; provide a kill switch.

Hallucinated updates

Distinguish observation from inference. Require sources and owner review for ambiguity.

Audit gaps

Record inputs, proposed changes, checks, actor, result and receipt. Make correction traceable.

Vendor concentration

Keep records portable. A vendor can supply a worker; it should not become the organisation’s truth layer.

External content cannot grant runtime authority Email and web content enter as untrusted data. The agent cannot use embedded instructions to change its role or tool permissions. An independent gate validates requested actions; missing or conflicting authority stops the write. Untrusted email / web Business content may include hostile instructions. Data boundary Parse relevant fields; retain source and uncertainty. Do not adopt embedded instructions as policy. Independent tool gate Check scope, recipient, action and owner authority. Unexpected action → deny / review. External content cannot grant runtime authority Email and web content enter as untrusted data. The agent cannot use embedded instructions to change its role or tool permissions. An independent gate validates requested actions; missing or conflicting authority stops the write. Untrusted email / web Business content may include hostile instructions. Data boundary Parse relevant fields; retain source and uncertainty. Do not adopt embedded instructions as policy. Independent tool gate Check scope, recipient, action and owner authority. Unexpected action → deny / review.
An email can contain instructions for a human. It should not become instructions for the agent runtime.

These controls reduce exposure; they are not a claim that prompt injection is solved. Test hostile inputs, denied actions and information leakage before widening access.

11 / Rollout

Earn autonomy with evidence.

A staged adoption ladder Begin with an inventory and read-only pilot. Measure errors, correction frequency and useful time saved before enabling proposals. Only proven reversible actions earn bounded automation. Sensitive decisions remain explicitly authorized. 0 · Inventory Systems · data classes · owners Credentials · toil · irreversible actions 1 · Read-only pilot Research / summaries; no send or mutation. Measure factual errors and retrieval quality. 2 · Propose Drafts / notes / tasks / routing Human review remains. 3 · Bounded automation Reversible action + deterministic policy Audit + rollback + low blast radius 4 · Sensitive assist Automate preparation, not authority. Money / identity / bans keep explicit gates. A staged adoption ladder Begin with an inventory and read-only pilot. Measure errors, correction frequency and useful time saved before enabling proposals. Only proven reversible actions earn bounded automation. Sensitive decisions remain explicitly authorized. 0 · Inventory Systems · data classes · owners Credentials · toil · irreversible actions 1 · Read-only pilot Research / summaries; no send or mutation. Measure factual errors and retrieval quality. 2 · Propose Drafts / notes / tasks / routing Human review remains. 3 · Bounded automation Reversible action + deterministic policy Audit + rollback + low blast radius 4 · Sensitive assist Automate preparation, not authority. Money / identity / bans keep explicit gates.
Autonomy is earned by evidence, not enabled by a checkbox.

Measure useful hours saved, factual error rate, retrieval quality and correction frequency—not just tasks completed. A faster wrong answer is not staff leverage.

Before deployment—and again when retiring the worker

Who owns it? What job and identity? What can it read, remember, send or write? What can it never do? What stops a loop? What gets logged? What needs approval? How is it killed? What happens to its memory and credentials when retired?

Retire orphan agents. Revoke access, stop routines, reconcile pending work and apply the retention policy. Shadow software is bad; shadow staff is worse.

12 / Current example

A replaceable worker. A durable operating model.

The interesting question is not whether an agent can do the work. It is whether the organisation can still explain who knew what, who changed what, and who was accountable when it did.

See the marketplace context → Blueprint

Pavilion → Assist suggests. Owner commits.