HOW IT WORKS
The capability behind the signal
An agent receives organisational context, plans a task, selects approved tools and performs bounded actions across applications. Role-based access, audit logs, approval gates and monitoring constrain what it may do.
EVIDENCE BASE
What supports the assessment
- 01
Workspace agents can be shared, scheduled and connected to approved tools.
- 02
Role-based permissions, audit logs and approval checkpoints are explicit product features.
- 03
The Tertiary Education Commission is procuring enterprise-grade generative AI access and capability building for potential use across approximately 420,000 tertiary staff and students; participation and uptake are not guaranteed.
- 04
AWS reports that Halter’s Clank agent triages system alerts and has saved more than 215 hours of manual engineering work; this is company and vendor evidence rather than an independent audit.
- 05
Stats NZ publishes 13 approved AI uses that are operating or moving toward implementation and three time-limited experimental trials, with stated safeguards and human oversight.
- 06
The Commerce Commission says existing competition and consumer law still applies when AI influences decisions and outcomes, and boards should have visibility of organisational AI use and safeguards.
GLOBAL INDUSTRY IMPACT
What could it change?
Agents are developing from personal assistants into governed operating infrastructure for recurring knowledge work across sales, IT, finance, research and operations.
NEW ZEALAND RELEVANCE
Why it matters here
New Zealand evidence now spans procurement intent, a deployed operational workflow, public transparency and an explicit regulatory accountability baseline. The next test is repeatable, independently measurable value and comparable disclosure across organisations.
IMPLICATIONS FOR NZ TECHNOLOGY
What it means for the sector
- Opportunity for local workflow products built on global models
- Demand for agent security, evaluation and audit tooling
- A larger integration market for SMEs and regulated sectors
- Need for New Zealand-specific legal, cultural and public-sector controls
PARTICIPATION PATHS
What New Zealand can do
- Choose one repeatable workflow with a clear owner
- Require approval before external or consequential actions
- Measure cost and success per completed workflow
- Build sector-specific connectors, evaluations and audit records
- Give boards a current inventory of AI use, affected decisions and consumer or competition safeguards
LIMITS & RISKS
What should not be overstated
- Prompt injection and unsafe tool use
- Errors compound across long tasks
- Access to business data increases security exposure
- Vendor-reported outcomes need independent validation
- Compliance responsibility cannot be delegated to a model or vendor
NEXT SIGNALS
What Digital.nz will watch
- Reliability on scheduled workflows
- Enterprise security and audit evidence
- Cost per completed workflow
- Tertiary-sector supplier selection and uptake
- Independent evidence from New Zealand deployments
- Comparable public-sector AI transparency registers
- Enforcement and case evidence under existing law