AI Agents for Landscape Architects: Design Revisions and Approvals
An AI agent for a landscape architecture firm tracks design revisions, client sign-offs and planting changes. Here is the approval record to keep.

An AI agent for a landscape architecture firm is most useful as the keeper of the project record: which plan revision the client approved, which planting list was in force on a given date, and who signed off on each substitution. It does not design the garden. It holds the job's memory, drafts status updates, routes each decision to the person who owns it, and treats a phone call without a written record as an open item. Firms that start here should begin with one approval record for every planting change, because that record is where disputes, wrong orders and rework usually begin.
Key takeaways
- A revision exists only when the approved version is named, dated and linked to the drawing set it replaced.
- Planting changes need their own approval trail, separate from general design sign-off.
- An agent should draft, route and remind. The designer of record and the client make the decisions.
- Every approval needs a source, a decision, an owner and a status that someone can check months later.
- Start with one project type and one approval path, then widen the scope after the record holds up.
Why revisions break down on landscape projects
A landscape project has more hands on it than most design work. The client, the developer, an arborist, an irrigation contractor, a nursery, and sometimes a municipal reviewer each touch the plan. Each of them asks for changes, usually in a different channel. A substitution that starts as a sentence in a site meeting can reach the nursery order before the drawing set is updated, and the planting schedule may still show the original species months later.
The memory of the job also tends to be scattered. Part of it lives in email, part in a marked-up PDF, part in a foreman's notebook, and part in the memory of whoever attended the walk-through. When a question comes up about why a bed holds a different shrub than the plan shows, the answer has to be reconstructed from fragments. That reconstruction is slow, and it is often wrong in the places where the client cares most.
An agent with access to the right records can keep those fragments connected. The principle is simple: every change has a reason, a proposer, a decision and a version it lands in. The agent's job is to make that chain visible and to stop it from being skipped.
What a project agent would track across revisions
A well-scoped project agent keeps a running model of the job rather than a pile of messages. For each project, it would track the following:
- The revision index, with the date, the trigger for the revision and the sheets that changed.
- Open comments from the client, the developer and reviewers, and which ones have been answered.
- Which version each stakeholder approved, in what form (signed markup, email, meeting minutes) and when.
- Site constraints that must survive every revision, such as root zone clearances, mature spreads, setbacks, and irrigation zone boundaries.
- Pending decisions, their owners and the date by which the project will be affected if no answer arrives.
- Supplier and nursery dependencies, including availability windows that shape what can still be changed.
The agent reads and writes records in these areas. It should not change the drawing set, edit the planting schedule on its own, or decide whether a substitute is horticulturally acceptable. Those actions belong to people with professional responsibility for the design. The agent makes sure they happen, in order, with a record.
The approval record for every planting change
The most valuable single artifact is a structured approval record for each planting change. Keep it in one place, linked to the revision it creates. A complete record contains:
- A change identifier and the exact location on the plan, such as a bed name, zone, or tree tag.
- The original species, cultivar, container or caliper size, and quantity.
- The proposed replacement, with the reason for the change, such as availability, soil conditions, maintenance concerns, or a preference expressed by the client.
- The horticultural check completed by the designer of record, covering mature size, light, water needs, and clearance from structures and utilities.
- The client decision: approved, approved with conditions, or rejected, with the date and the channel used.
- A link or reference to the evidence, such as the signed markup or the message in which the client wrote the approval.
- The name of the person who recorded the decision, and whether an agent drafted the entry.
- The revision in which the change appears, and the planting schedule version that reflects it.
- The supplier confirmation, when the order depends on the change.
A record like this turns a vague memory into a fact that anyone on the team can verify. It also gives the agent something concrete to check before it releases a status update that says a bed is ready to order.
A planting change without an approval record is still an open change, even if everyone agrees on it.
Step by step: putting a gate around a substitution
Illustrative scenario: a courtyard planting plan specifies a particular shrub. The nursery reports that the cultivar will not be available in the spring window. The site contractor suggests a substitute during a call with the project manager. Here is how a project agent would handle the change under a clear approval gate:
- Step 1: The agent opens a change record with the original species, the proposed substitute, the reason given, and the person who raised it.
- Step 2: The agent checks the planting schedule and the irrigation zone map, then flags every bed and zone the substitute would touch.
- Step 3: The agent drafts a short comparison note for the designer of record, covering mature size, water needs and light. It does not send this to the client.
- Step 4: The designer confirms or rejects the horticultural fit and records the decision in the change record.
- Step 5: If the designer accepts the substitute, the agent sends the client a decision request that states the exact change, the revision it will create, and the date by which an answer is needed to keep the spring order.
- Step 6: When the client approves in writing, the agent creates the new revision and attaches the approval with its channel and time.
- Step 7: The nursery order is released only after the approval record exists and is linked to the revision. If the client has not answered, the agent flags the delay to the project manager rather than substituting silently.
Notice what the agent does not do. It does not accept a phone approval as final, and it does not change the schedule because a contractor was confident. Its value is in keeping the sequence honest.
Where permissions and approval boundaries belong
An agent that can send messages, edit files and place orders needs clear permission limits. The general guidance on authorization is useful here: grant each tool the narrowest access needed for the task, separate read access from write access, and keep final decisions with people who hold the authority to make them. For a landscape firm, that usually means the agent can read drawing metadata and the approval log, write draft notes and change records, and send requests through approved channels. It cannot approve on behalf of the client, cannot alter the drawing set, and cannot release a supplier order that lacks a linked approval.
The context an agent uses also needs boundaries. Project memory should be scoped to the job at hand, so that a substitution on one site is not mixed up with a similar plant on another. The Anthropic engineering guidance on context engineering describes the general problem of giving an agent the right information at the right time, without flooding it with everything the firm has ever written.
In a Be Human engagement, the first step defines one role, such as a revision and approval tracker for a single project type, along with the memory sources it may read, the tools and permissions it holds, the approval boundaries it must respect, and an observable success criterion. Managed operations then cover monitoring of the record, recurring evaluations against finished projects, permission reviews when people or tools change, and controlled updates to the workflow. The client keeps authority over every design and client decision throughout.
How to check that the approval record is working
A practical test is to take a finished project and trace one planting change from first proposal to installed plant. Can you find the request, the horticultural check, the client decision, the revision it created, and the order that followed? If any link is missing, the gate has a hole, and the agent's workflow needs adjustment before it handles live jobs.
Over time, a few simple signals show whether the record is healthy: how many planting decisions remain open past their needed-by date, how many orders were released without a linked approval, and how often a revision had to be reissued because a change was not recorded. The goal is not a dashboard for its own sake. It is an early warning when the record starts to drift from the work on site. The NIST AI Risk Management Framework describes this kind of ongoing measurement and governance as part of responsible AI use, and the OpenTelemetry observability primer explains why logs and traces that show what happened and when are the basis for reconstructing any system's behavior after the fact.
Firms that already keep submittal or RFI logs will recognize the pattern. The approach used for engineering project memory and for RFI and submittal logs applies here as well: one record per decision, a named owner, and a visible status.
If you want to find the first revision or approval task worth handing to an agent, the free Be Human AI diagnostic is a practical place to start.
Frequently asked questions
Can an AI agent approve a planting change for a client? No. The agent can draft the request, attach the horticultural notes, and track whether an answer has arrived. The approval itself must come from the client through a channel the firm has defined, and the agent should record it, not produce it.
What should a landscape firm store for each planting change? Store the original and proposed species with sizes and quantities, the reason for the change, the designer's horticultural check, the client decision with date and channel, the revision it created, and the supplier confirmation if an order depends on it. A record missing any of these items leaves a gap that someone will have to fill from memory later.
Does a firm need a large system to start? No. A firm can begin with one project type, one structured log, and one agent role that reads the log and drafts updates. The useful test is whether every planting change in a live job can be traced from request to installed plant, and that test works in a spreadsheet or a project tool the team already uses.
How does this relate to permissions and security? An agent that can write to project records or send client messages should operate under the narrowest access that its task requires, with clear separation between drafting and deciding. Review those permissions when people, tools or workflows change, and keep an audit trail of what the agent read, wrote and sent so that the firm can explain any decision afterward.
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.


