Superintelligence Prep for Marine Service Yards: Haul-Outs and Parts
A practical guide for marine service yards on AI agents for haul-out scheduling, parts status and owner updates, and the records to build now.

An AI for a boat repair shop works best when it takes on the repetitive coordination around the work: keeping haul-out slots current, reporting parts status accurately and answering owners with confirmed information. Decisions about quotes, extra work, delivery dates and release stay with the people who run the yard. The records a yard builds now, such as stable boat identifiers, dated parts events and written approvals, decide how much a more capable agent can safely do later.
Key takeaways
- Start with haul-out scheduling and parts status, because both generate repeated calls and both have states that can be checked against a record.
- Give any agent a written list of what it may say and what only a person can approve, including quotes, extra work, substitutions and delivery dates.
- Identify every boat by a stable record that includes hull and engine details, so two vessels are never matched by owner name alone.
- Log parts as dated events (ordered, promised, received, fitted) rather than as a single status word that nobody can audit.
- Keep approvals in writing, with the exact wording the owner accepted, because that record defines what an agent may repeat later.
Why haul-outs and parts are the right place to start
A haul-out depends on tide windows, travel lift or trailer availability, stand and cradle space, weather and the order in which boats are worked. A parts order depends on supplier timing, whether the fitting in stock matches the engine serial, and whether a substitution has been approved. Owners ask the same few questions again and again: when is my boat coming out, is the part here yet, when can I launch. Each answer depends on several records that often live in different places, such as a whiteboard in the shop, a supplier email and a note from the yard foreman.
This is a good starting point because most of the work is retrieval, comparison and drafting. An agent can read the records, notice that a promised date has passed, and draft a clear message for a person to review. The decisions that carry financial or safety weight remain with the shop. That split keeps the first agent useful without asking it to make judgments the yard has not yet written down.
Map the states a boat moves through
Before any agent touches the schedule, write down every state a boat can occupy in the yard. Most yards use some version of the list below, and each state needs one source of truth and one named person responsible for changing it.
- Booked for haul-out
- Waiting for lift or stand space
- Hauled and on stands
- Waiting for parts or owner approval
- Work in progress
- Ready to splash
- Released and settled
The transitions matter as much as the states. A boat should not move from waiting for approval to work in progress unless the approval is recorded with its date and wording. If the yard currently moves boats by verbal agreement on the floor, the first task is to write that agreement into the record, not to automate it. An agent that reads an out-of-date board will repeat the error faster than a person would.
Build a parts record that can be checked
Each part should be a line item tied to one boat, with a supplier reference, an ordered date, a promised date, a received date, a fitted date and a note on any substitution. The discipline is in recording events as they happen. A status field that says "on order" tells an owner nothing about whether the supplier has confirmed the promise, and an agent cannot tell whether that label is recent.
When an owner asks about a part, the agent should report only what the record shows: the last confirmed event, its date, and whether the promised date has passed. It should never convert an expected delivery into a launch date. If a substitution is proposed, the agent may describe the difference in the record, but a person must approve it in writing before the part is fitted.
Illustrative scenario: An owner asks whether the replacement raw water impeller has arrived. The record shows an order placed earlier in the week, a promised date that has already passed and no received entry. A well-configured agent replies that the part is not yet logged as received, that the supplier's promised date has passed, and that the yard manager will confirm a new date that day. It does not promise a launch date, and it flags the stale entry for the parts lead.
How should owner updates stay accurate?
Owner updates should be built from the record and say clearly what is confirmed and what is expected. A useful update names the boat, states the current state in plain words, gives the date of the last confirmed event and names the next action and who owns it. Photos should come from the job card for that specific boat and task, not from a general gallery that can be mixed up.
The words "should be ready" are a common source of trouble. They sound reassuring, but they become a commitment in the owner's mind. An agent should avoid them unless a staff member has confirmed the date in the record. Where the answer is uncertain, the update should say that the yard is checking and when the owner will hear back. Clear uncertainty is easier to manage than a date that later moves.
In a managed engagement, Be Human would begin by reviewing how haul-out slots, parts events and owner messages are recorded today. From there the team would define one or two narrow tasks for the agent, write the approval boundaries into its operating instructions, and set up monitoring so the yard can see what was sent, what was read and when a person stepped in. Permission review and incident response continue after launch, so access to the scheduling calendar, supplier mailbox and customer channel stays limited and is checked on a schedule.
Approval boundaries for a marine service yard
Approval boundaries should be written as plain sentences that a new employee could follow. They should also be enforced by the access the agent has, not only by its instructions. Authorization guidance from OWASP is a useful reference here, because it treats access as something to check against a specific action on a specific resource rather than against a general role. An agent that can read the whole parts ledger does not need permission to edit it.
- The agent may propose an open haul-out slot, but a person confirms it on the schedule.
- The agent may report the last logged status of a part, but it may not invent or estimate a delivery date.
- The agent may draft a message about additional work, but only the service writer sends a quote or approval request.
- The agent may not confirm a release, change a payment arrangement or waive a fee.
- The agent may not accept a substitution part without written sign-off from the responsible person.
- The agent escalates any owner message that mentions damage, injury, insurance or a dispute to a named person.
The last rule matters more than it seems. Owners sometimes describe damage in the same conversation where they ask about a launch date. An agent that is configured to pass those messages to a person early protects both the owner and the yard.
Records to build now for a more capable agent
Today's agent needs accurate inputs. A more capable agent later will need richer memory, longer context and possibly wider access to tools. The yard cannot make that future safe by adding access later. It has to build the records now, so that every new capability reads from the same accountable source. The OpenTelemetry observability primer describes logs, metrics and traces as ways to understand what a system did and why. A yard can apply the same thinking to its agent: what it read, what it sent, which rule it followed and when a person intervened.
The records worth building are specific. Each boat needs a stable identifier, hull and engine details, and a contact with a clear record of which channels the owner has agreed to. Each approval needs a date, the exact words accepted and the person who recorded it. Each parts event needs the supplier reference and the date it occurred. Haul-out history should show each lift, stand move and splash, with the reason for any change. Photos should be tied to the job they document. Decisions should carry a short reason, so that the next reviewer is not guessing why a substitution was accepted.
An agent cannot infer intent from a pile of messages. It can only use what the yard has recorded clearly. That is the reason to build these records before any capability expansion, not after.
A weekly review that keeps the agent accurate
The first months of an agent in a yard should include a short weekly review. The review is not a long audit. It is a checklist that keeps the record and the floor aligned, and it gives the yard a way to notice drift before an owner does.
- Compare every boat's recorded state with the shop floor board and correct any mismatch.
- List parts whose promised dates have passed without a received entry, and assign each one to a person.
- Read every outbound owner message the agent sent that week and note any that should have been escalated.
- Check the access the agent has to the calendar, the supplier mailbox and the customer channel, and remove anything it no longer needs.
- Record each decision made from the review, with its reason, so the next review starts from the same facts.
If you want to find the first task in your yard that is worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Do we need new software before an AI agent can help in the yard? Not always. Many yards start with the records they already keep, such as a shared schedule, a parts list and a job card, and tidy them into consistent states and dated events. The agent reads from those records and drafts messages for review. New software can come later if the existing tools cannot hold the information the agent needs.
What should an AI agent never say to an owner? It should never promise a launch date, confirm a delivery that the record does not show, approve extra work, accept a substitution or settle a payment question. It should also avoid phrases like "should be ready" unless a staff member has confirmed the date in the record. When a message touches damage, injury, insurance or a dispute, the agent should pass it to a named person right away.
What happens when a boat's parts record is incomplete? The agent should say what is known and what is not, using the last confirmed event and its date. It should flag the gap to the parts lead, not fill it with an estimate. Completing the record is a human task, and the weekly review is the place to catch gaps before they become owner complaints.
Is this a good first step for a yard without a dedicated operations person? It can be, provided the first task is narrow and the approval boundaries are written down. A yard without an operations role often benefits from a coordinator that keeps the schedule and parts status consistent, while the owner or service manager keeps the decisions. The key is to start with one workflow, review its output each week and expand only when the records have proven reliable.
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.


