What a Day With an AI Agent Looks Like in a Septic Design Office

An AI agent in a septic design office takes the repeatable parts of a job, such as intake, drafting notes, file cleanup, package assembly and follow-up, and leaves every engineering and regulatory decision to you. A realistic day looks less like a robot designing systems and more like a careful assistant who has the drafts ready when you sit down and tells you plainly what it could not finish.

Key takeaways

• The agent drafts and organizes. The designer of record reviews, decides and signs.
• The biggest gains show up in the gaps between design tasks: intake, chasing missing information, formatting and packaging.
• A good agent reports what it did, what it skipped and what it is unsure about, in plain language.
• Access should be narrow: specific folders, specific tasks, no ability to submit anything on its own.
• Start with one repeated task and prove it on a job you already know before adding a second.

What does an AI agent actually do for a septic designer?

Think of the agent as a staff member who never designs. It reads what arrives, files it where it belongs, prepares first drafts of the paperwork you write over and over, and keeps a list of what is open. When it is unsure, it asks. When something is outside its task list, it stops.

The Anthropic guide on building effective agents makes a point that applies directly here: simple, well defined workflows usually beat open ended autonomy. A septic office is a good fit for that advice, because the tasks repeat and the stakes of the decision are high. You want a narrow tool with a short, checkable output, not a free roaming system.

The tasks that tend to fit are intake from a new request, organizing the survey base and the soil data, producing a first pass of the design notes, assembling the permit package from your approved sheets, and following up on missing items. The tasks that do not fit are choosing the system type, sizing, placing the field, and judging whether a site is suitable. Those are yours, and your state rules frame them. For Texas, that framework is 30 TAC Chapter 285, linked from the TCEQ OSSF program page.

What does one day look like?

This is an illustrative day for a one-person office with a handful of active jobs. It does not describe a real client or a measured result.

• Early morning: the agent checks the shared inbox and the job folders. It lists new requests, jobs waiting on a missing item and anything that came in overnight, in a short list at the top of your day.
• Intake: for each new request, it creates the job folder, records the address and the parcel details the client gave, and drafts the questions you usually ask about the house, the bedrooms count, the water source and access to the site.
• Base preparation: for a job you have already scheduled, it gathers the county data you pointed it to and prepares the base drawing for you to review, flagging anything that did not line up.
• Midday: you do the design work. The agent stays out of the way except to answer questions about what is in the folder.
• Notes drafting: from your inputs, it fills the parametric notes template and marks every field that it could not fill, so you see the gaps first.
• Afternoon package: once you approve the sheets, it assembles the PDF package in the order your permit office expects, runs a completeness checklist and tells you what is missing.
• End of day: it writes a short log of what it touched, what it left open and what needs your attention tomorrow.

Notice what is missing from the list. There is no step where the agent chooses the system or signs anything. Each task produces an output you can review in a few minutes.

How do you set the boundaries so it stays safe?

Boundaries matter more than features. The OWASP authorization guidance describes the principle of least privilege: give a system only the access it needs for the task. In a design office, that translates into practical rules.

• Folder access is limited to active job folders and the templates folder.
• The agent can read the shared inbox but cannot send an email to a client or the permit office without your approval.
• It never uploads to a permit portal. You do the submission.
• Every output is a draft with a visible label, and your review changes the label.
• Client personal data stays in the job folder and is not pasted into general tools.
• You keep a log you can read, so you can see what it did on any job.

The NIST AI Risk Management Framework is a useful reference when you want to describe these controls to a partner, an insurer or your own checklist: map where the agent is used, measure how it performs on your tasks, and manage the risks you find.

What should you hand over first?

Pick the task you repeat most and dislike most. For many designers that is the first draft of the design notes or the assembly of the package, since both are formatting heavy and rule based. Others start with intake because a missed question costs a second site visit.

A simple way to choose:

• List the tasks you did on your last three jobs.
• Mark the ones that were identical each time except for the details.
• Cross out anything that required a judgment about the site or the system.
• Pick one from what remains, and write down what a correct output looks like.
• Run it on a finished job and compare the agent's draft to what you produced.

If the draft matches in structure and the gaps are flagged honestly, you have a task to keep. If it does not, adjust the instructions, or drop it. You are not committing to anything by testing.

How does Be Human run this?

Be Human designs, implements and operates personalized AI agents for small businesses. For a septic office, we start by watching how a job moves through your day, then build one narrow agent for the first task, with the access limits and the draft labels described above. We operate it afterwards: we monitor it, adjust it when your templates or your local rules change, and keep the log readable. The work is remote, in English and in Portuguese. More about how we run agents is on the managed AI agents page.

Several pieces of this day have their own guides. For the CAD side, see AI for the AutoCAD site plan. For the paperwork, see the permit package routine, and for a broader look at the question of whether this fits your office, read should a septic designer use an AI agent.

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

Frequently asked questions

Will an AI agent replace the septic designer? No. The designer of record carries the professional responsibility, makes the engineering and regulatory decisions and signs the package. The agent handles drafting, organizing and reminders so your time goes to the parts only you can do.

Can the agent talk to my clients directly? It can draft messages, but sending should stay behind your approval. Client communication carries your name and your license, so a draft that you read and send is a better default than automatic replies.

How much setup does a one-person office need? Much less than people expect, if you begin with one task. You need your templates, a folder structure and a few finished jobs to test against. Adding a second task is easier once the first one is stable.

What happens when the agent is wrong? It will be wrong sometimes, which is why every output is a labeled draft and the log shows what it touched. You catch errors in review, and the correction goes back into the instructions so the same mistake does not repeat.

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