Journal
What "agentic engineering" actually means (and what it doesn't)
Agentic Engineers · 4 min read
- Category
- Operations
A short definition of the category we're claiming. What agentic engineering is, what it isn't, and why we chose to make it a noun.
We are trademarking "Agentic Engineering" in Australia, so it is worth saying plainly what we mean by it before anyone else gets to define it for us. A category is either defined deliberately by the firm that built it, or accidentally by whichever vendor uses the term loudest in the first eighteen months. We intend the first.
Agentic engineering is not a tool stack. It is not "the way we use ChatGPT internally." It is not prompt engineering, not a dev shop with Copilot turned on, and not an AI-themed pitch deck with the same workflow underneath. The term has been borrowed by enough vendors who mean none of those things that the definition needs to be staked out clearly.
What it is
Agentic engineering is an operating system for shipping software. Agents handle the operational layer. Humans handle judgment, product, and review. The agents take the repetitive friction off the team. The humans take the work that compounds. The system as a whole moves faster than the sum of its people, and faster than any single operator could move regardless of how leveraged they are individually.
Three properties matter, in this order, and any definition that drops one of them is describing something else.
- It compounds across engagements. Anything we build for one client gets sharper for the next. The Sprint we ran six weeks ago made this Sprint cheaper. The review automation we tuned on one client's repo catches a class of bug on every client's repo. This is not how traditional dev agencies work, where each engagement starts at zero and the only thing that travels between clients is whichever operator happens to roll over. Compounding across engagements is the entire reason the category is worth defining.
- The system is the differentiator, not the individual. Every agency claims to staff brilliant engineers. The honest answer is that any single engineer is a single point of failure. We sell the operating system. Engineers move through it. The system is what stays. If a single operator is the unit, it is staffing, not agentic engineering.
- Discipline lives on top of the agentic layer. AI without discipline is faster slop. Agentic engineering is slow on the things that should be slow (architecture, contract design, tiering decisions, monthly close) and fast on the things that should be fast (boilerplate, sprint reporting, code review first pass, recruitment screening, ticket triage). The discipline is the part that takes years to install. The agentic layer is the part that takes a Sprint.
AI without discipline is faster slop.
What it is not
It is not the use of AI tools by an individual engineer. An operator with a strong Copilot setup is using AI tools well. They are not running agentic engineering. The unit is the team, not the seat.
It is not a single product. There is no agentic-engineering SaaS to buy. The category is an operating model installed inside a firm that ships software for clients, not a tool the buyer plugs in.
It is not a re-brand of "AI-augmented development." Augmentation describes an additive layer on top of an unchanged workflow. Agentic engineering describes a workflow that has been restructured around the assumption that agents do the predictable work. The restructuring is the substance.
It is not a posture. Posture is what the marketing department prints on the homepage. Agentic engineering is what runs in CI, in the review queue, in the monthly close, in the recruitment funnel. If you cannot see it in the operating tools, it is not there.
Verbs are activities. Nouns are categories. "We do agentic engineering" is a sentence any vendor could say. "We are an agentic engineering firm" makes a claim about what the firm is, not about what the firm does this week. The trademark is the public commitment to defending the category against the watering-down that always happens once a term gets adopted by the rest of the market.
Why the definition matters now
Two reasons, both timing-related.
The first is that the category is in the window where it gets defined. In any new category, the first eighteen months are when the definition is shaped by whoever writes about it most clearly. If we leave that to the rest of the market, the term will be diluted into "we use AI" inside a year, the same way "DevOps" was diluted into "we have a CI pipeline" by 2018.
The second is that buyers are starting to ask the right question. The question used to be "do you use AI." It is becoming "where does the AI actually run." That is the question we built the firm to answer, and the firm only works as an answer if the category is defined sharply enough that the answer is testable.
If you are building a SaaS company that needs to move from 1 to N, that is the conversation we are here for. Book a Sprint and see how the system runs against your stack.
