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

AI by industry

Superintelligence Prep for Concrete Crews: Pour Schedules and Tickets

Pour schedules and delivery tickets can become shared memory for a concrete crew. See how an AI agent reasons over them and where people approve.

Be Human typographic cover with the title: Superintelligence Prep for Concrete Crews: Pour Schedules and Tickets
Conceptual illustration created with AI · Be Human

AI agents for concrete contractors are most useful when they reason over the records a crew already keeps: the pour schedule, the batch and delivery tickets, and the notes that tie them to a placement. With that shared memory in one place, an agent can flag a truck arriving outside the placement window, a ticket with no job reference, or a pour that now needs a different finishing crew. The agent prepares the answer and the draft message, while the foreman, the plant contact and the special inspector keep the decisions.

Key takeaways

  • A pour schedule is already a memory system. The real work is making it consistent enough for one agent and for the people who sign off.
  • Delivery tickets carry facts the schedule cannot hold, such as truck identity, batch time, discharge time and any water added on site.
  • Start with reconciliation and exception flags, not with dispatch changes or quoting decisions.
  • Scope every tool to read or draft first. Anything that changes an order, a placement or a hold point needs a named person to approve it.
  • Measure success by whether exceptions reach a person before the next truck leaves, not by how much the agent writes.

Why a pour schedule is already a memory system

A pour schedule looks like a calendar, but on a working site it holds much more. For each placement it records the element, the mix specified, the planned volume, the pump or chute setup, the crew, the weather window and the hold points that must be passed before concrete goes in. Most crews keep this in a whiteboard photo, a spreadsheet, a dispatch tool or a foreman's phone, and those pieces rarely agree by the end of the week.

Treating the schedule as memory means each entry becomes a stable reference. A placement has an identifier, a set of expected deliveries and a responsible person. When a truck is rerouted or a placement moves a day because of rain, that change should update the record instead of living only in a text thread. An agent can only reason about a pour if it can find the pour, and it finds it reliably only when the record carries a clear identifier that the delivery side uses too.

What the delivery ticket adds that the schedule cannot

The delivery ticket is the second half of the record. The plant or dispatch system generates it for each load, and it typically states the ticket number, the truck, the mix, the quantity, the batch time and, once the truck arrives, the discharge time. Many crews also log slump checks, air content and any water added at the site, because an addition changes the mix and must be documented. Those are the facts an inspector or engineer will later ask about.

Read on their own, tickets are a pile of receipts. Read against the schedule, they answer useful questions: did every planned truck arrive, how long did the interval between batching and discharge run, which loads are not attached to any placement, and which placement has more concrete on order than its layout needs. The useful work is the join between the two sources, and a capable agent handles that join well when both sources use the same placement identifier.

Set up the shared record before the agent touches it

Work through these steps in order. Each one removes a category of exception the agent would otherwise have to explain later.

  • Give every placement one identifier that appears on the pour schedule, the work order and the delivery ticket reference.
  • Name the element, not only the zone, so a footing, a wall and a slab on the same level are never confused.
  • Record the specified mix and its allowed tolerance in one document that the responsible people sign off, not in a note on a truck.
  • Log hold points as separate items, each with a named person who passes or fails it and the time it was done.
  • Decide who may change a placement time, and write that rule next to the schedule.
  • Connect or export tickets daily so the schedule and the ticket log are never more than one working day apart.
  • Keep a short written note for each change: what moved, why, and who approved it.

Notice what the list does not contain. No step asks the agent to decide anything. The first job is to make the record trustworthy, because an agent layered over a wrong record will be confidently wrong, and that is worse than no agent at all.

What a capable agent can reason over, and what it should not decide

A capable model can read a schedule and a set of tickets, compare them, summarize the differences and explain them in plain language. It can also hold the context of the week: which placements depend on which hold points, which finishing crews were booked, and which trucks were recently rerouted. That context is the reason for the shared memory. It is not a reason to hand over the outcome of a pour.

Use this decision guide when you define the agent's role:

  • If a ticket has no placement identifier, or one that matches no placement, the agent flags it to dispatch for correction and does not guess a match.
  • If a load arrives outside its planned window, the agent drafts a notice to the foreman and the plant contact. A person decides whether to accept, hold or reject the load.
  • If water appears on a ticket or a logged slump falls outside the specified range, the agent escalates to the responsible person and the inspector. It does not propose a change to the mix.
  • If a placement moves, the agent drafts an updated schedule and lists every affected ticket. The foreman approves the change before it is written to the shared record.
  • If the record is silent on something, the agent says so. Missing data is an exception to report, not an assumption to fill in.

Illustrative scenario: a slab pour on a Tuesday

The following scenario is illustrative and does not describe a real client. A crew has a slab placement scheduled for early morning with six loads planned. By the time the crew arrives, five tickets already carry the placement identifier, and one ticket has a typo in the job reference. Mid-morning, the ticket feed shows the fourth truck batched earlier than planned but still outside the window the foreman set for discharge, and the schedule shows the reinforcement hold point as not yet passed.

The agent sends the foreman one message with three items: the fourth load is late against the window, the sixth ticket is unattached, and the reinforcement hold point is open. The foreman calls the plant to confirm the load, asks the inspector to check the reinforcement before placement, and tells the agent to close the flags once the checks are done. The agent writes a note to the record with each action and its time. The agent decided nothing. It made the gaps visible while there was still time to act.

Where approvals and inspections stay with people

Agent access should follow the same logic as any crew member's access. Read access to the schedule and the ticket log is reasonable. Write access should be limited to draft notes and a task queue. The agent should not change plant orders, edit specified mixes or mark a hold point as passed. Access control guidance such as the [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) is a useful checklist when you write these permissions.

Subcontractor notes, scanned tickets and free-text comments are also untrusted input. A note that says to ignore the hold point should never change what the agent does. The [OWASP prompt injection prevention cheat sheet](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) describes the general defenses. Finally, specification compliance, sampling and acceptance remain governed by the project documents and the people named in them. Software organizes the evidence; it does not certify the concrete.

Be Human starts with one role, usually reconciliation and exception reporting for pours. Before anything is switched on, the team defines the memory the agent may read, the tools it may call, the approval points and the success criterion. Managed operations then cover monitoring of the ticket feed, recurring checks that the agent's flags still match how placements are actually run, integration maintenance when dispatch software changes, permission reviews, and controlled updates to the agent's instructions, each with a written record. The service is remote, and the crew keeps the decisions.

A related example for a trade with similar field records is the irrigation contractor's zones and repairs approach, which uses the same principle of a stable identifier joined to every visit.

Where to start with the first pour

Choose a small, bounded first use rather than a full dispatch rebuild. A practical sequence looks like this:

  • Pick one active pour next week and rebuild its schedule and ticket record by hand, noting every mismatch.
  • Sort the mismatches by who can fix each one. Many will be dispatch corrections, not agent problems.
  • Choose one exception type, such as tickets without a placement identifier, and decide whether a daily flag would help.
  • Write the approval rule for that exception before any tool is connected.

For a crew that already keeps careful records, the same method applies to a landscaping team's routes and bids, where memory has to come before agents. Be sure the first task is one your people already trust. The free Be Human AI diagnostic can help you find the first task worth handing to an agent.

Frequently asked questions

Can an AI agent read delivery tickets from our batch plant? It can, but only if the plant or dispatch system exports tickets or exposes them through an interface your crew is permitted to use. Confirm the data terms with the supplier and check which fields actually appear on the ticket before connecting anything. Start with read-only access and a daily export.

Does an agent replace the special inspector or the concrete technician? No. Hold points, sampling and acceptance stay with the people and the engineer of record named in the project documents. The agent organizes evidence and reminds the responsible person of open items, but it does not certify placements.

How long does setup take? Setup depends on how many systems the record spans and how consistent the existing placement identifiers are. A first engagement defines one useful role, the memory and tools it needs, the approval boundaries and an observable success criterion before anything is operated. Crews with clean schedules usually have less cleanup to do first.

Is this only useful for large contractors? The same logic applies to a small crew with one foreman and a shared spreadsheet, as long as placement identifiers and hold points are written down. Be Human delivers this work remotely in English and Portuguese, so the size of the crew matters less than the discipline of the record.

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 →