Agentic Operating Model
A short engagement that designs how AI assistants and your team actually work together. Using the Edison Autonomy Ladder (assisted → copilot → autopilot → self-driving) to grade every agent. So when you implement AI agents, they fit your business rather than disrupt it.
The pattern we keep seeing.
A business commissions a lead-qualifier agent in March. Ops sets up a status agent in April. Marketing buys a content agent in May. Nobody can name what's running where. The first time someone asks 'who owns the customer-facing chatbot?' is also the first time the chatbot says something embarrassing. The agents arrived faster than the operating model.
Agents are appearing one workflow at a time, with no map.
Sales bought a lead-qualifier. Ops set up a status agent. Marketing has a content agent. Nobody can name what's running where, or who reviews what.
Approval gates are inconsistent.
Some agents send emails directly. Some need manager review. Some need legal sign-off nobody documented. The pattern feels improvised because it is.
Risk is creeping.
Without an operating model, the first time an agent does something embarrassing in front of a customer is also the first time anyone thinks about governance.
What is Agentic Operating Model?
The documented system that describes how AI assistants and human staff share work. Who owns what, who approves what, what runs automatically and what stays human. The engagement that prevents agentic AI from becoming agentic chaos.
A short engagement that designs how AI assistants and your team actually work together. Using the Edison Autonomy Ladder (assisted → copilot → autopilot → self-driving) to grade every agent. So when you implement AI agents, they fit your business rather than disrupt it.
Edison AI designs agentic operating models for Australian SMBs. The documented system describing how AI assistants and human staff share work, where humans approve agent actions, what's automated and what stays human. Standard outputs include an agent roster, an approval gate map, a human-in-the-loop standard and a governance one-pager. Engagement runs 3–4 weeks, $12,000–$30,000 plus GST. The Edison Autonomy Ladder (assisted → copilot → autopilot → self-driving) is used to grade every agent.
The shifts you can't postpone.
Three reasons the operating model is cheap before agents are live and expensive after.
- 01
Agentic AI is moving from pilot to production.
The next 12 months will see most Australian SMBs running at least one agent. The teams that win run them governed, not improvised.
- 02
Customers and boards are catching up to the question.
'Does AI ever act on our behalf without a human checking?' is now a procurement and board-level question. The right answer is a one-pager, not a 20-minute explanation.
- 03
Designing later is twice the work.
Retrofitting governance to running agents costs more in both consulting and internal disruption than designing the model first. The cheap option is the early one.
What you get.
- 01
Current-state agent inventory
- 02
Autonomy Ladder mapping (per agent)
- 03
Agent roster (owner, scope, data access, approval gates)
- 04
Approval gate map (visual)
- 05
Human-in-the-loop standard
- 06
Governance one-pager
Where this shows up.
Sales agents
Autonomy
Typically copilot. AI drafts; the rep approves and sends.
Gate
Outbound message review before send.
Owner / data scope
Head of sales owns. Data scope: inbound CRM, lead enrichment, prior interactions.
Operations agents
Autonomy
Autopilot for internal status; copilot for external supplier comms.
Gate
External messages reviewed; internal summaries auto-posted.
Owner / data scope
COO or operations lead. Data scope: project tool, status records, supplier registers.
Support agents
Autonomy
Copilot for first-response drafting; autopilot for ticket routing.
Gate
Agent review before send for first responses; auto-routing on classification.
Owner / data scope
Head of support. Data scope: helpdesk records, knowledge base, customer history.
Knowledge agents
Autonomy
Autopilot inside the business; reviewed for externally surfaced answers.
Gate
Internal: none. External: human review before publish.
Owner / data scope
Head of operations or knowledge lead. Data scope: policies, SOPs, product docs, internal wiki.
Reporting agents
Autonomy
Autopilot (read-only across data sources).
Gate
Anomalies above threshold escalate to the named owner; otherwise auto-publish.
Owner / data scope
Finance or operations lead. Data scope: KPI dashboard, transactional systems, BI tools.
Customer-facing agents
Autonomy
Tightly governed. Typically copilot with pre-approved response templates.
Gate
Response templates pre-approved; weekly QA sampling; documented escalation path.
Owner / data scope
Head of customer + head of product. Data scope: only the public knowledge base and approved customer data.
The engagement.
- Step 01
Diagnose
Inventory current and planned agents. Identify governance gaps. Interview the function leads who will own agents.
- Step 02
Design
Map each agent to the Edison Autonomy Ladder. Define ownership, data scope, approval gates and escalation paths.
- Step 03
Deploy
Draft the roster, the gate map and the human-in-the-loop standard. Run the team training session so operators know how to work inside the model.
- Step 04
Embed
Governance one-pager handed over. Quarterly review cadence set. Optional fractional oversight for businesses with active agent rollouts.
What changes.
Every agent has a named owner, scope and gate.
No more 'who runs that?' across the leadership team. The roster is the single answer. Written, owned, reviewable.
Approval gates documented and consistent.
What happens automatically and what needs a human is written, visible and reviewable in a single page. The board can read it; the auditor can read it; the team can follow it.
One governance one-pager that closes procurement objections.
Removes the dominant 2026 procurement objection in a single PDF. Customer questionnaires, board governance review and audit checklists answered from the same document.
Who this works for.
This is for you if…
- You have one or more agents already running and no written governance
- You're about to commission an AI agent build and want the model designed first
- You're being asked about AI governance by customers, board or auditors
- You want a written standard your team can follow without consulting Edison every time
- You want the operating model that lets you scale agents without scaling risk
Not the right fit yet if…
- You have no current or planned agents (start with playbooks and foundations)
- You want a deep enterprise-grade governance review with IRAP / PSPF certification (we'll refer)
How this compares.
Five common alternatives to a designed operating model. Only one is sized for an Australian SMB and built around how operators actually work.
No operating model
- Gives
- Speed in the short term
- Falls short
- Risk creep, hidden agents, surprises in front of customers
- Edison difference
- Lightweight model designed in 3–4 weeks
Copy a Gartner framework
- Gives
- Comprehensive theory
- Falls short
- Doesn't match SMB reality. Written for enterprises with 500+ staff
- Edison difference
- Sized for $1M–$50M revenue businesses
Build it after the agents are live
- Gives
- 'We'll catch up'
- Falls short
- Retrofitting governance is twice the work and twice the cost
- Edison difference
- Design before or alongside, never after
Big consultancy operating model
- Gives
- Brand-credible deliverable
- Falls short
- Six figures, slow, over-engineered for SMB use
- Edison difference
- Boutique, fixed-fee, practical
Internal IT writes one
- Gives
- Familiar to IT
- Falls short
- Technical, not operational. Written for the architect, not the operator
- Edison difference
- Built for the operators who run the workflows day-to-day
Edison AI
Operator-grade, founder-led, fixed quote. Built around your real stack and workflows , not a binder, a brochure, or a six-figure off-the-shelf programme.
What buyers ask first.
“We're too small for an operating model.”
Operating models scale with the business. The Edison version is built for $1M–$50M businesses. A one-pager, not a binder. The point is to make governance reviewable, not consultative.
“Will this slow down our agent rollout?”
No. The operating model is the artefact that lets agents go faster, not slower. Because nobody has to argue about approvals as you go. Decisions are made once, written down, and reviewed quarterly.
“Doesn't this overlap with responsible AI training?”
They complement. Responsible AI sets the safety rules for the team. The operating model sets the rules for the agents. Many clients run both. Usually responsible AI first, then the operating model.
Common questions.
What's the investment range for an agentic operating model engagement?
$12,000–$30,000 plus GST depending on the number of agents in scope and the depth of governance work.
How long does it take?
3–4 weeks end-to-end. Most engagements run a 1-week diagnostic, 1–2 weeks design, then deploy and embed in the final week.
Do we need to have agents running already?
No. Many engagements happen before any agents are commissioned. That's the cleanest sequence. The model becomes the design brief for the agents that follow.
What if we already have a governance framework?
We'll fit Edison's Autonomy Ladder and roster into your existing framework rather than replacing it. The operating model becomes a layer, not a duplicate.
Who needs to be involved?
A sponsor (COO, GM or founder), the function leads who'll own agents, and IT if the data scope crosses sensitive systems.
Will this satisfy customer questionnaires on AI governance?
Yes, for the vast majority of mid-market and lower-enterprise vendor questionnaires that ask about AI governance, approval gates and human oversight.
Does this lock us into specific tools?
No. The model is platform-agnostic. We design the rules; you choose the tools.
What's the relationship with Bespoke AI Systems?
This engagement is the strategy. The bespoke build engagement is the execution. Many clients run them sequentially. Operating model first, build second.
Other training services.
- S01
AI Workshops for Teams
A practical AI workshop for Australian SMB teams. Half-day or full-day, on your real work, building one shared prompting standard the whole team can follow.
Learn more - S02
Role-Based AI Workflow Training
Function-by-function AI workshops that rebuild your team's most-repeated workflow with AI, so the work runs faster, looks the same every time, and doesn't depend on one person.
Learn more - S03
Executive AI Enablement
Confidential 1:1 and small-group AI coaching for Australian founders, CEOs, directors and leadership teams, on the work that actually fills your week.
Learn more - S04
AI Playbooks & Prompt Systems
A short project that turns your team's AI know-how into a library of prompts and one-page playbooks, so new hires inherit it on day one and capability doesn't leave when staff change.
Learn more - S05
Cybersecurity & Responsible AI Training
A practical training programme that gives your team the rules, the habits and the documentation to use AI safely, without banning the tools that are actually helping.
Learn more - S06
AI Adoption Programme
A structured six-week AI adoption programme that combines team foundations, two role-based workflow rebuilds, a productised prompt library, a responsible-use policy and fortnightly leadership check-ins.
Learn more