Workflow agent
A workflow agent is an AI system that moves one business process from start to finish with explicit scope, guardrails, an evaluation harness, and a defined path to escalate cases it cannot handle. It changes how a whole process runs, not how one person works.
A workflow agent takes a supplier invoice from inbox to accounting system with the GL code mapped, or a patient intake form to a structured clinical note, without a person re-typing the data in between. This is the thing Core Nova builds in an AI Workflow Enablement engagement. It is a scoped build, not a chat session and not a licence.
Per-seat assistant
A per-seat assistant is a general AI tool licensed to an individual to help them write, summarise, and search faster, billed per user per month. It makes a person quicker; it does not change how a business process runs.
Microsoft 365 Copilot and ChatGPT Team are per-seat assistants. They are a useful productivity floor, but most Australian SMEs that bought seats have nothing operational to show for the spend, because the savings live in the workflow, not the seat. If you have paid for assistants and nothing has changed on your P&L, the next step is a workflow agent, not more seats.
First Workflow
First Workflow is Core Nova's fixed-price entry engagement: A$6,000 + GST, four working days of senior practitioner time, with the first measurable improvement by week two. You leave with a process map, one workflow shipped to production, and a runbook, and you keep the result either way.
The engagement maps your business end to end, identifies the process losing the most recoverable hours, and ships one improvement before the second week closes. It ends there unless both sides agree there is more value to capture. First Workflow is the public entry to the AI Workflow Enablement practice, not a standalone product.
Scope drift
Scope drift is when an AI agent gradually starts handling work it was never designed or tested for, because no one set a hard boundary on what it is allowed to do. It is one of the three failure modes most AI consultancies do not name.
An agent built to draft customer replies starts quietly making refund decisions; an agent built to file documents starts interpreting them. Core Nova documents the scope at the end of Day 1 of a First Workflow and builds the boundary into the agent, so the work it does is the work it was tested on.
Escalation gap
An escalation gap is when an AI agent silently handles a case that should have gone to a person, because no rule was set for what it must hand off. The damage is invisible until someone finds the case the agent got wrong.
The fix is an explicit escalation path: defined conditions under which the agent stops and routes the case to a human instead of guessing. Core Nova defines escalation gates for every workflow it ships, so the agent's confidence does not become your liability.
Evaluation rot
Evaluation rot is when an AI agent's output quality degrades over time as the underlying data, formats, or business rules shift, and no one notices because nothing is measuring it. The agent keeps running; the answers quietly get worse.
An agent tuned to one report template starts mishandling cases when the template changes, or a classifier drifts as the input mix moves. The defence is an evaluation harness that keeps checking the agent against a known-good standard. Without measurement, the first sign of evaluation rot is usually a client complaint.
Evaluation harness
An evaluation harness is the test setup that measures whether an AI agent is still producing correct output, run against a fixed set of known-good cases. It is what catches evaluation rot before a customer does.
The harness runs the agent against an eval set and flags when results fall below the agreed standard, the same way a test suite catches a code regression. Every Core Nova workflow ships with an evaluation method built in, not bolted on later. An agent without an evaluation harness is an agent no one is actually checking.
Eval set
An eval set is the curated collection of real cases, each with a known correct answer, that an AI agent is tested against to prove it still works. It is the reference standard an evaluation harness measures against.
A good eval set covers the normal cases and the edge cases that matter to your business, so the test reflects real work rather than a demo. Core Nova builds the eval set during the engagement and hands it over in the runbook, so the standard the agent was held to stays with you.
Governance scaffold
A governance scaffold is the set of controls, documentation, and accountability that sits around an AI deployment so it can be governed, audited, and explained. Core Nova builds it against NIST AI RMF and ISO/IEC 42001.
It gives a board, an auditor, or an insurer something concrete to read: what the agent does, what data it touches, where it escalates, and how its output is checked. The scaffold sits on top of the workflow, not in front of it, so governance does not become the reason nothing ships. It is the same security and compliance lens a vCISO engagement brings, applied to AI.
Agent programme
An agent programme is an ongoing engagement that deploys and maintains multiple workflow agents across a business, with shared governance and continuing evaluation. It is the shape an engagement can grow into after a First Workflow has proven its value.
Where First Workflow ships one improvement, an agent programme covers a set of workflows under one governance scaffold, with evaluation harnesses kept current as the business changes. Core Nova scopes a programme only after the first improvement ships, never before. The conversation starts with one workflow that pays for itself.
Ready to put one of these to work? Book a 30-minute scoping call.
