Agentic Infrastructure Design: The New Discipline That Will Reshape How We Design

There is a discipline forming as I write, and it deserves a name of its own. I call it Agentic Infrastructure Design or AID. This is my attempt to define it clearly — because if we are going to build the future of civil engineering on top of it, we should at least agree on what it is.

What Agentic Infrastructure Design Actually Is

Here is the definition I use:

Agentic Infrastructure Design is the process of designing civil infrastructure by directing an AI agent — or a team of specialised sub-agents — in natural language, rather than operating legacy software by hand.

Read that again, because the shift is subtle but total.

Today, an engineer designs by operating software. You open the legacy software like CIVIL3D, InfoDrainage, Revit, etc. You click through menus. You enter values, run the model, read the output, change the values, and run it again. The design lives in your head; the software is just the tool you use to externalise it. The skill being tested is not really the engineering judgement — it is your fluency with the software interface.

Agentic Infrastructure Design inverts that. You stop operating the software and start directing an agent. You state the requirements. The agent does the work autonomously— sizing, modelling, iterating, and checking against engineering standards. You review, and sign off a validated, traceable design.

Describe it. The agent designs it. You review and approve it.

Why This Is Not "AI-Assisted Design"

I want to draw a hard line here, because the phrase "AI in design" has been stretched to mean almost anything.

A chatbot bolted onto your legacy CAD software (like C3D AI assistant) is not agentic design. An autocomplete that finishes your spreadsheet formula is not agentic design. A tool that generates a report from data you already produced is not agentic design. Those are co-pilots — helpful, incremental, and fundamentally still leaving you at the controls.

Agentic Infrastructure Design is different in one specific way: the agent does the design work autonomously, not you. It performs the actual engineering task — the catchment delineation, the network layout, the stormwater control sizing, the compliance checks — and it does so through mix of high-level reasoning, deterministic gates and intelligent tool calling (ITC).

The engineer's hands come off the operating software. That is the threshold. Everything before it is assistance. Everything after it is agentic.

Why Now, And Not Five Years Ago

The technology readiness is only one half of the story. The other half is that the industry cannot afford not to change.

Consider the pressure building on UK infrastructure alone:

• Schedule 3 of the Flood and Water Management Act is making SuDS mandatory, and with roughly 1.5 million new homes planned, each one now needs a compliant sustainable drainage design.
• The £718 billion UK infrastructure pipeline over the coming decade is real work that has to be delivered, with construction output already rising.
• And we are trying to deliver it into a talent gap: the sector needs around one million more engineers by 2030, while an estimated 20% of the current workforce retires within five years.

Put those three facts side by side and the conclusion is uncomfortable but clear: the demand is rising faster than our capacity to meet it. We cannot hire our way out. We cannot work longer hours out. The only lever left is a step-change in how much a single engineer can produce — and that is precisely what agentic design offers.

Meanwhile, drainage design still eats the week. Manual catchment delineation. Tedious iteration on stormwater control sizing. No-surcharge and no-flood checks. Endless back-and-forth inside expensive legacy software. The work is essential and the work is slow — and slow is the one thing we no longer have room for.

The Trust Question: Could You Stamp Your Name To It?

Whenever I present this, an experienced engineer asks the question that actually matters. Not necessarily "can it work?" — but "could I put my name to it?"

It is the right question. And Agentic Infrastructure Design is only worth anything if the answer is yes. So the discipline has to be built around trust from the first principle, not bolted on afterwards. Four things make that possible:

• Human-in-the-loop. The agent iterates and designs. The engineer reviews and approves. Nothing goes out the door without a qualified professional signing it off. The authority — and the accountability — stays exactly where it belongs.
• Dual validation (drainage example). No-surcharge and no-flood checks are built into the process, not left as an afterthought. The design is checked autonomously by a checking agent as it is created.
• Standards-traceable. Every output can be traced back to the source: the relevant Code, the CIRIA C753 SuDS Manual, the applicable regulatory thresholds. A design you cannot trace is a design you cannot defend, and the agent has to show its working (full transparency).
• Augmentation, not replacement. This is the one I feel most strongly about. The goal is not to remove the engineer. It is to free the engineer from the operating software so they can spend their time on high-level judgement & taste — the part of the job that actually requires a human decision that will make an impact on the environment.

Same standards. Same accountability. A fraction of the manual effort.

From Weeks To Minutes

This is not theoretical. It is what we build.

With Floabl, we brought agentic capability directly into the InfoDrainage workflow that engineers already use — so the intelligence meets you where you work. And with RUUTE AI, we are going further: designing compliant drainage from scratch, in plain English, from a blank site to a full compliant & validated network.

What used to take weeks of manual effort can be produced in minutes — with the same standards applied and the same accountability preserved. The engineer moves from doing the work to directing and validating the work. The output is not a shortcut. It is a validated, traceable design that a professional can confidently sign.

What the Future Looks Like

Drainage today. Infrastructure tomorrow.

Drainage is where we PERVAN AI are starting, because it is a domain of structured procedures, clear standards, and repeatable checks. But the discipline does not stop at drainage. Agentic highways design is coming next to RUUTE. Then structures. No more multiple expensive & complex design software for each sector. Do you get our vision?

And I think the trajectory of the discipline itself unfolds in three phases:

1. Phase 1 — Human-in-the-loop (where we are today). AI agents design. The engineer reviews and approves every output.
2. Phase 2 — Human-on-the-loop. An AI Master Reviewer (much capable and reliable than the human) checks each design. The engineer signs off and focuses on stakeholders.
3. Phase 3 — Human-out-of-the-loop. Agents liaise, submit, and sign off, while the client's own agentic system reviews the work in detail.

Across all three phases, the progress trajectory is the same: less operating software, more engineering judgement for humans.

The Skill That Will Matter Most

Here is the part I find most exciting, and it is where I want to leave you.

When an agent can design any civil infrastructure, the engineer's core question shifts. It moves away from "how do I design this?" — the mechanical question that has consumed most of our careers — toward something far more interesting: "should this be designed at all? Should it even exist?"

In a world where the how is handled, the scarce, valuable, irreplaceable human skill becomes taste. Judgement. Knowing what is worth designing & building, for whom, and why. Knowing when the elegant answer is to build less, or nothing at all. Taste becomes the greatest engineering skill.

That is not a diminished role for the engineer. It is a promotion. We spent thirty years learning to operate the software. That era is coming to an end. The next thirty years belong to fully autonomous agents — while we humans focus solely on taste and on sharpening our high-level judgement.