Skip to content
be-human.ai
← Back to the journal

AI by industry

Superintelligence Prep for Security Firms: Post Orders and Shift Logs

How security guard firms turn post orders and shift logs into structured memory so an AI agent can answer site questions without changing approved procedures.

Be Human typographic cover with the title: Superintelligence Prep for Security Firms: Post Orders and Shift Logs
Conceptual illustration created with AI · Be Human

An AI agent for a security guard company can answer site questions reliably when post orders and shift logs are converted into structured, versioned memory, with a named owner for every rule. The agent reads only approved procedures and the logged facts from recent shifts, states which version of a procedure it used, and sends any request that would change a procedure to a supervisor. It does not grant access, rewrite a post order, or edit a log entry after it was written.

Key takeaways

  • Post orders are the procedure of record. The agent should quote them and never reinterpret them.
  • Shift logs are evidence. Store each entry as a time-stamped event with an author, and correct it only by appending a new entry.
  • Every post order needs a version, an effective date, an approver and a site scope before it enters memory.
  • The agent answers questions and flags gaps. Changes to procedures, access decisions and dispatch stay with people.
  • Start with one site and one question type, test the answers against the approved version, then widen access.

Why post orders make the right first memory

Post orders are the written instructions a guard follows at a specific post: who may enter, which doors are checked, how often rounds happen, what counts as an alarm, and whom to call in what order. They already exist, an accountable person approves them, and they change slowly. That combination makes them a stable base for an agent, which is why they are a better starting point than inbound messages or dispatch traffic.

An agent that has to guess the current rule creates risk for the firm. An agent that can point to the rule, its version and its approver is useful to a shift supervisor, a client representative and a new officer working their first night. The value is not the chat interface. It is the speed at which a correct, sourced answer can be found.

Most firms keep post orders in several places: a binder at the desk, a PDF attached to the client contract, and an amendment that was sent by email and never filed. The first task is therefore not technical. Someone has to decide which document is the current version, mark the others as retired, and keep the retired copies for audit. Until that decision is made, no memory built on top of the documents can be trusted.

What belongs in a post order record

When post orders are loaded into memory, each rule should be a separate record rather than a page of prose. The original document stays as the source, but the agent retrieves smaller units with metadata attached. A useful record carries the following:

  • Site and post identifier, so a question about the north gate cannot pull the loading dock rule.
  • Procedure section, such as access control, patrol, alarm response, visitor escort or end of shift.
  • Rule text copied word for word from the approved document, with the page or section reference.
  • Version number, effective date and the name of the person who approved it.
  • Approval class: informational, requires supervisor review, or requires client sign-off.
  • Conditions, such as holidays, events or temporary closures, written as explicit conditions instead of implied exceptions.

The last item matters most. Post orders often contain phrases like unless otherwise directed. A model cannot resolve that phrase on its own. The record should name who gives the direction and where that direction is logged. If the direction is not recorded anywhere, the agent should answer that the exception needs a supervisor.

Turning shift logs into structured events

Shift logs describe what happened. They contain timestamps, completed rounds, visitor names, vehicle plates, equipment faults, handover notes and sometimes a short narrative written at the end of a long night. Logs are valuable for questions such as whether the east fence was checked before an alarm was raised. They also contain personal data and occasionally incomplete or inaccurate statements, so the memory must treat them with care. The same discipline used for document-heavy operations, such as the process server service record approach, applies here: the record must show who wrote what and when.

Convert each entry into an event with a fixed set of fields: site, post, time, author, event type, free text and a reference to any attachment. Event types should come from a short controlled list, for example round completed, visitor logged, equipment fault, incident reported, handover and client instruction received. Keep the original free text. The structure makes retrieval precise, while the free text preserves what the officer actually wrote.

Corrections follow the same rule. If an entry was wrong, a new correction event points to the original, and the original remains visible. An agent answering from a log must say which entry it used and whether a correction exists. Overwriting history would make every later answer unverifiable, which is the opposite of what a supervisor needs during a review.

A step-by-step plan for the first memory build

A first build should be small enough to test completely. The following sequence keeps the scope narrow and gives the owner a clear point to approve or reject each stage:

  • Choose one client site that has a current, signed set of post orders and a reasonable history of shift logs.
  • Name an owner on the firm side who can approve changes to that site's post orders and who will answer questions about them.
  • Collect every version of each post order, mark the one in force, and move retired versions out of the working folder while keeping them for audit.
  • Split each post order into rule records using the fields above. Have the owner review a sample of records for accuracy before anything is loaded.
  • Define the controlled event types for shift logs and map existing entries to them. Notes that do not fit become free text.
  • Write the questions the agent must answer from real calls, such as which access list applies on a holiday or what the patrol interval is after the building closes.
  • Set the answer format: the rule, its version, its approval class, and its source reference. If no approved rule matches, the agent says so and names the supervisor.
  • Test with the owner before any officer uses the system. Record every answer that was wrong, incomplete or overconfident, and fix the memory before widening access.
  • Review the memory on a fixed schedule, covering new post order versions, retired rules and questions the agent could not answer.

Where the agent must stop

The boundary between helpful and unsafe is easier to hold when it is written down. The following guide is a practical starting point for the owner to adapt:

  • Answering what the post order says is in scope, with the version cited.
  • Answering what happened on the previous shift is in scope for users who are authorized to read that site's logs.
  • Drafting a proposed change to a post order is acceptable as a request routed to the owner. The agent never publishes the change itself.
  • Granting, extending or changing access for a person, vehicle or visitor is outside scope. Those decisions belong to the client or supervisor under the firm's permission rules.
  • Choosing a dispatch or incident response is outside scope. The agent may show the escalation list written in the post order and nothing beyond it.
  • Editing or deleting a log entry is outside scope. Corrections are appended by the author or a supervisor.

Logs also create an input risk. A free-text note could contain sentences that try to change how the agent behaves, whether typed by mistake or on purpose. Retrieved text should be handled as data to be summarized, not as instructions to follow. The OWASP guidance on prompt injection prevention covers this class of problem. Access should follow the principle of least privilege described in OWASP's authorization guidance, so a user sees only the sites and log types their role requires.

Illustrative scenario: a night officer asks whether a contractor with a valid visitor pass may enter the loading dock after the gate closes. The agent retrieves the access rule for the dock, which says after-hours deliveries require a call to the duty supervisor. It also finds a note from the previous shift about a scheduled delivery. It answers with the rule, its version and the log reference, and states that it cannot approve entry. The officer calls the supervisor, who decides. The agent did not make the access decision, and the answer can be traced later.

How a managed approach keeps procedures under human control

Be Human designs and operates this kind of memory as part of a managed service. A first engagement defines one useful role for the agent, the post order and log memory that role needs, the approval boundaries, and an observable success criterion, such as whether answers match the approved version when the owner tests them. Ongoing work can include monitoring, recurring evaluations, permission review, integration maintenance, incident response and controlled changes when a post order is revised. The firm keeps authority over its procedures, and the agent keeps the record easy to query. For a broader view of how shared memory is governed across operations, see managed service provider memory.

If you want to find the first task worth handing to an agent in your operation, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent tell a guard what to do during an alarm? It can show the escalation steps written in the approved alarm response section and cite the version used. It should not improvise steps or decide whether a situation counts as an incident. Dispatch and judgment remain with trained staff and the supervisor on duty.

Do we need to rewrite our post orders before an agent can use them? Usually not, but the documents need to be split into rule records with versions, owners and effective dates. Ambiguous phrases such as unless otherwise directed should be clarified by the owner, because the agent cannot resolve them reliably on its own.

Is it safe to put shift logs into an AI memory? It can be, if access follows need-to-know rules, retrieved log text is treated as data rather than instructions, and corrections are appended instead of overwritten. Logs often contain personal data about visitors and staff, so the obligations depend on where you operate and should be reviewed with qualified counsel.

How do we know the agent is answering correctly? Write a set of real questions with the approved answers, test the agent against them, and review every failure with the post order owner. Repeat the test after each post order revision and on a regular schedule, so drift and missing rules are caught before an officer depends on them during a shift.

Your next step

See where an AI agent fits in your company first.

Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →