Managed AI governance & agent care
Agents degrade as the business changes. This is the layer that notices.
What it is
The durable layer after go-live: ongoing review of agent output, governance and responsible-AI check-ins, performance tuning as your processes drift, and new agent builds as needs emerge. It exists because the failure mode of a deployed agent is rarely a crash — it is a slow decline nobody is watching for.
Who it is for. Organisations running agents in production that cannot afford to have them quietly get worse.
The steps, in order
The same sequence every time, so you can tell where an engagement is up to without asking. How the phases fit together across a full programme is set out on our approach.
Review the output monthly
A sample of each agent's real output is reviewed against the same evaluation set used at build time.
Watch for drift and exceptions
Exception rates, rejected outputs and human overrides are tracked over time — the early signal that a process has changed underneath an agent.
Check in on governance quarterly
Agent inventory, risk register and human-oversight rules revisited against what actually changed in the business.
Re-score annually
A yearly re-run of the readiness benchmark, so governance posture is tracked rather than assumed.
Absorb small changes
An allocation of change time each month for the tweaks that would otherwise queue behind a new statement of work.
Build new agents as needed
New agents scoped and built by arrangement, using the delivery pattern already proven in your environment.
What you walk away with
Named deliverables, handed over as artefacts you own and can use without us.
| Deliverable | What it contains |
|---|---|
| Monthly output review | A short report on how each agent performed against its evaluation set, with anything that moved called out. |
| Exception and override log | What humans corrected and how often — the most honest measure of whether an agent is still earning its keep. |
| Quarterly governance pack | Updated agent inventory, risk register and oversight rules, ready to hand to a board or an auditor. |
| Annual benchmark re-score | Your readiness score re-run each year, with the change from last year explained. |
| Monthly change allocation | A named allocation of hours for small changes and tuning, so minor fixes do not need a new engagement. |
How long it takes, and what it costs
Ongoing, month to month
Elapsed time from kick-off to handover, assuming your people are available for the sessions the steps above describe.
$6K – $10K Per month
Indicative only. The retainer is not open for booking yet, and the band will be confirmed when it opens.
All price bands on this site are indicative ranges in AUD, ex-GST. The final fixed price is quoted per engagement once scope is agreed. Software licences (Claude, Microsoft 365 Copilot or another enterprise stack) are client-owned and passed through at vendor pricing — never bundled into our fee.
What it depends on, and what comes next
AI deployment & agent build
Agents running in production. There is nothing to care for before that.
AI deployment & agent build →This is the steady state
Nothing after this — the retainer is the steady state, month to month, cancellable.
Not open for booking yet
Available from 2027. Agents we deploy today ship with monitoring, an evaluation set and a run-book so your own team can review their output; the managed service that takes that work off your hands opens in 2027.