Before you trust an agentic system, test it - with your own people.
Who checks your agentic system - someone who didn't build it? Eight principles for deploying AI agents, alone or in more complex agentic systems, each with a test you can run inside your own firm - so you know where your AI agents inform decisions about people.
The free Manifesto Kit: the 2026 Manifesto PDF with an evidence test under every principle, the 10-minute self-assessment scorecard, the 2026 preface, the Agent Evidence Index spreadsheet to start your own inventory, and a scorecard worksheet with a worked example.
PDF by email. Marketing email only if you tick the box.
AI transparency for firms using agentic systems: Australia and the EU →
Name the person who checks an agentic system they didn't build.
If it's the person who built it, that is self-assessment - step one, not evidence. Principle 08: verified trust over self-assessment. The PDF has a test under every principle and a 10-minute scorecard.
Augmentation over Automation
Bounded Adaptability over Rigid Programming
Transparency over Black-Box Operation
Human-Centred Design over Technical Efficiency
Accountable Collaboration over Isolated Operation
Responsible Operation over Unrestricted Capabilities
Paced Improvement over Static Deployment
Verified Trust over Self-Assessment
Two regimes, one starting point: know where your agents make or inform decisions about people. Our guide to Australia's automated-decision transparency rules and the EU AI Act starts with that inventory, side by side. Read the guide → or get the Manifesto Kit (free) →
Start free. Go as far as your clients, insurer or regulator will one day ask.
Every offer applies the same eight principles. Start where your firm is.
The Manifesto Kit
The 2026 Manifesto with an evidence test under every principle, the 10-minute scorecard, the 2026 preface, the Agent Evidence Index and a scorecard worksheet with a worked example.
Get the Manifesto Kit (free) →Agentic Institute*
Foundations, self-paced, and the Evaluator Workshop, half a day in-house for your developer and an evaluator. The Health Check for the agentic systems you run. Both start from your Agent Evidence Index and Evidence Trace - required to enter. EU AI Act training for staff and boards.
Work with George
AI Strategy & Advisory
Decide what to build, buy or leave alone - with reasons your partners can defend. The Brief has a fixed price. Everything after it is priced only once scope, design and timeline are agreed.
Fractional Chief AI Officer
Dr George Vossos joins your leadership team on a monthly cadence and owns the AI agenda: strategy, governance, vendors and delivery.
AI Software Development
When advice turns into a build, we design and ship production AI systems in collaboration with our engineering partner Evoke AI.
* Not accreditation, not a regulatory or ISO/IEC 42001 certification, not an EU AI Act conformity assessment, and not legal advice.
AI transparency: what Australia's privacy rules and the EU AI Act mean for the agents you run →
Your vendor's course, governance e-learning, or nothing?
| Option | What it gives you | What it doesn't |
|---|---|---|
| Free vendor AI courses | Good grounding in one vendor's tools | Independent evidence about your own deployment |
| AI governance credentials and e-learning | Individual knowledge of policy and regulation | Tests of how your agents actually behave |
| Monitoring and observability tools | Whether it's running, and what it did and why - usually read by the team that built it | Someone who didn't build it checking that the evidence is enough - and that people can stop it |
| Doing nothing | No cost today | Anything to show a client, insurer or regulator who asks |
Vendors teach you to drive their car. The Manifesto teaches you to inspect any car.
The four layers
Netlifestyle AI's governance pattern - not part of the Manifesto, not legal advice
- MonitoringIs it running?
- ObservabilityWhat did it do, and why?
- TraceabilityCan someone else follow it back?
- The ManifestoIs that enough evidence - and can they stop it?
Each layer needs the one before it, and none replaces the one after. None of them is a safeguard on its own: the safeguards are the hand-back, the limits and the stop (Principles 02, 06 and 05). The evidence shows whether they exist - and whether they fire.