AI Agent for Scenic Shops: Quotes, Build Schedules and Change Orders

An AI agent for a scenic shop drafts quotes from the drawing package, keeps each change order attached to the drawing revision it modifies, and warns the shop when a build task is falling behind the date the piece must be ready for load-in. It works from the shop's own records through permissions the owner sets. The estimator and the shop foreman still decide what leaves the building.

Key takeaways

• A drawing-based quote needs a fixed checklist of materials, labor stages and finish work, so the agent can show how each line was built.
• Every change order should name one piece, one drawing revision, a requester and an approval state before any cost or date moves.
• Schedule warnings are useful only when they compare build progress with the fixed load-in and tech dates, not with internal targets nobody outside the shop sees.
• Pricing, accepting changes and committing dates to a production office should stay with named people.
• Start with one role, such as quote drafting for flats and platforms, and judge it against a standard you can check.

Why scenic quotes drift away from the drawings

Scenic work is a moving target. A designer sends elevations, a revised sheet arrives after the first estimate, a director asks for a wider opening during tech, and a paint elevation changes once a sample is approved. Each change touches materials, labor and the calendar at the same time. When these changes live in scattered email threads and marked-up PDFs, the quote can end up describing a set that no longer exists.

Illustrative scenario: a shop quotes a three-section platform unit from an early sheet. Two revisions later the platform height changes and a ramp is added. The estimator updates the lumber list but not the casters or the rigging points, so the quote and the build diverge by the time the crew starts cutting. Nobody made a careless mistake. The information simply lived in too many places.

This is the first problem an agent can address, because it can read every revision in the project log and flag which quote lines rest on a superseded sheet.

How the agent prices a flat, platform or set piece from a drawing

A useful quoting agent follows the same steps a careful estimator would, and it writes down each step so the estimator can check it.

• Read the drawing set, identify the current revision of each sheet, and flag any sheet older than the newest revision in the project log.
• Break each piece into components: frames, skins, bracing, deck boards, legs, hardware, rigging points, paint and texture, and transport.
• Map each component to the shop's standard build procedure and to the stock it normally uses, including sheet goods, dimensional lumber, fasteners and finish materials.
• Estimate crew hours for each stage from the shop's completed jobs, and mark any estimate that rests on a thin history.
• Assemble the quote with the assumptions written beside each line, so a number can be challenged without reopening the whole drawing.
• Hold the draft until a named estimator approves it. It should not reach the producer on its own.

The arithmetic is the part owners trust least, which is why the output should show its working. The estimator should be able to see which sheet produced which line, which assumption was used, and which lines are guesses. A quote that explains itself is easier to correct than one that only states a total.

For shops that buy much of their material from outside vendors, the quote often depends on supplier prices that change between estimate and order. The same pattern appears in purchasing work, which is covered in AI Agents for Procurement: Quotes and Supplier Follow-Up.

Tracking change orders against the drawing revision

Change orders are where scenic margins disappear. A request from a designer, a producer or a director should enter the system as a record that names the piece, the drawing it modifies, the requester, the estimated effect on materials, labor and schedule, and the current approval state. The agent's job is to keep that record complete and to show the before and after clearly.

• The request names one piece and one drawing sheet.
• The previous revision and the requested revision appear side by side.
• Cost and schedule effects are drafted, labeled as estimates, and tied to the specific request.
• The change stays pending until the authorized person approves it in writing.
• An approved change updates the quote, the cut list and the build calendar together, so they cannot disagree.
• Rejected or superseded changes remain in the log with the reason, so the same argument is not repeated at the next review.

A concrete rule helps here. No change that affects cost or a delivery date should move into the cut list until its approval is recorded. The agent can enforce that rule by refusing to update production documents from an unapproved request.

A change order that cannot show its previous revision is an opinion, not a record.

Building the fabrication schedule and warning when a build day slips

A scenic schedule has more dependencies than most whiteboards show. Design sign-off comes before cutting, cutting before assembly, assembly before paint, paint before drying, and drying before transport. Each stage depends on materials, a person and a bay. Load-in and tech dates are set by the production calendar, and an internal target is only useful when it is checked against that fixed point.

Illustrative scenario: the platform decks are scheduled for paint early in the week, but the steel brackets are still on backorder. The agent sees that the paint stage now finishes after the load-in buffer has been used up. It posts a warning to the foreman and the producer's coordinator, naming the affected pieces and the cause, before the crew has lost a day waiting on hardware.

Warnings need three things to be useful: a reason, an owner and a decision. A message that says a build is behind schedule is noise. A message that says the paint stage no longer fits before load-in, the bracket order is the cause, and the foreman must choose between reordering, reassigning the bay or asking the producer to move the truck is something a person can act on that day.

The discipline behind this resembles how software teams watch live systems. Each status change is recorded with a time and a source, so a late stage can be traced back to the event that caused it. The [OpenTelemetry observability primer](https://opentelemetry.io/docs/concepts/observability-primer/) describes this kind of signal-based view, and the same idea applies to a shop floor: if a stage slips, the record should show when and why.

What the agent should never decide on its own

Authority must be explicit before any agent touches quotes or schedules. In a typical scenic shop the roles are clear even if they are informal: the estimator approves quotes, the production manager accepts changes that carry cost or date impact, the foreman sets build sequence and bay use, and the owner sets margin floors. The agent can prepare, compare, remind and warn. It should not send a price to a client, accept a change, commit a date to a production office, or place or move a supplier order without the named approval.

Access should follow the same logic. The [OWASP authorization guidance](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html) describes the principle of granting only the access a function needs. For this agent that means read access to drawings and stock records, write access to draft quotes and the change log, and no write access to the supplier account or the client-facing email. Those permissions should be reviewed on a schedule, not left as they were set on the first day.

Text from outside the shop also needs care. A supplier email, a pasted designer note or a comment in a PDF can contain instructions that a language model may follow by mistake. Treat external text as data to be read and summarized, not as commands. The [NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) is a useful reference for this kind of work: name the risks, measure how the system behaves, and assign an owner to each control.

A decision guide for choosing the first task

Most shops do not need an agent for everything. The first role should be small enough to check and important enough to matter. Use these questions to choose.

• Does the task have a written standard you can check, such as a quote that must list every component and its source revision?
• Does it happen often enough that errors show up within a few jobs, not once a season?
• Is the expertise written down, or does it live only in a senior estimator's memory? If the latter, write the standard first.
• Is there one person who can approve or reject the output?
• Does the agent read more than it writes at the start?
• Does the task move money, commit dates to outside parties or change supplier orders? If so, defer it until the earlier tasks are stable.

A shop that answers yes to the first four questions can usually start with quote drafting or change order tracking. A shop that cannot yet answer them should spend its first effort writing its build standards, which is useful even without an agent.

How Be Human approaches a scenic shop engagement

Be Human starts with one defined role, often quote drafting or change order tracking, and specifies the memory and tools that role needs, the approval boundaries and an observable success criterion, such as whether every quote line traces to a named drawing sheet. Managed operations can then cover monitoring of the agent's outputs, recurring evaluations against a fixed set of past jobs, maintenance of the connections to the drawing and scheduling tools, periodic permission review, incident response when a warning is wrong, and controlled improvements made one change at a time. The work is remote and available in English and Portuguese. Shops with a deep record of past builds may also find the approach described in Superintelligent Agents for Engineering Firms: Project Memory relevant to how that history is organized.

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

Frequently asked questions

Can an AI agent price a scenic build from a drawing without an estimator? No, and it should not be set up that way. The agent can read the drawing set, break the piece into components, and draft a quote with its assumptions shown, but a named estimator should review and approve every quote before it reaches a producer or client.

What access does an AI agent need in a scenic shop? It needs read access to the current drawing revisions, stock records and past job data, plus write access only to draft quotes and the change log. It should not have authority over the supplier account, client email or production approvals, and those permissions should be reviewed regularly.

How does the agent warn that a build day is slipping? It compares each stage's expected finish with the fixed load-in and tech dates, using the dependencies between cutting, assembly, paint, drying and transport. When a stage no longer fits, it posts a warning that names the affected pieces, the cause and the decision that must be made, such as reordering hardware or reassigning a bay.

How should a shop judge whether the agent is working? Judge it against a standard that can be checked, such as whether each quote line traces to a named drawing sheet and whether warnings were correct when reviewed against what actually happened. Run recurring evaluations on a fixed set of past jobs, and change one thing at a time so the cause of any improvement or error stays visible.

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