- AI
- RUN
- C9 · MANAGED AGENT OPERATIONS
- 12-MONTH SUBSCRIPTION
Agents in production needoperations.Not just prompt engineering.
A UAE enterprise operating seven production agents across three business functions commissioned a 12-month Managed Agent Operations subscription. Per-active-agent commercial with monthly SLA cadence. Agent behaviour drift monitored continuously — three drift incidents surfaced and resolved in Year 1 before business impact materialised. Two agents recommended for retirement based on operational evidence, one new agent onboarded to production through the subscription's quarterly release cadence.
N=1 ILLUSTRATIVE COMPOSITE — representative pattern for a UAE enterprise operating at scale. Details drawn from patterns across NexITC engagements and market data. Not a specific client narrative.
THE SITUATION
The situation
The client is a UAE enterprise operating a portfolio of production AI agents deployed across the previous 18-24 months following separate build engagements. Portfolio composition at engagement start: seven agents across three business functions (customer service, operations support, internal knowledge worker productivity), built by three different implementation partners on two different underlying platforms.
The Chief Data Officer's team had inherited operational responsibility for the agent portfolio without a dedicated operational model. Individual agents had been running "in production" but without unified observability, without structured monitoring for behaviour drift, and without a defined process for handling agent output degradation short of full re-deployment. Two agents had exhibited noticeable output quality degradation over the preceding quarter — user-reported anecdotally but without measurement infrastructure to characterise the pattern.
The CDO's problem was structural. Individual prompt engineering fixes on degraded agents worked in the short term but did not scale as the portfolio grew. Model drift, data drift, and behaviour drift each required different diagnostic approaches, and the internal team did not have the operational discipline (or bandwidth) to run comprehensive drift monitoring across seven agents while continuing to develop new agents.
THE SPECIFIC QUESTION
The specific question
The clinic conversation surfaced what "structural" would mean. The CDO did not want a diagnostic report on the two degraded agents. The CDO wanted operational cadence — monthly SLA-measured agent-portfolio operations that held output quality continuously across all agents, with drift detection running as monitoring not as reactive investigation, and with a defined process for adjusting agent configuration or recommending retirement when operational evidence supported it.
The Practice Lead's scoping response confirmed C9 as the right shape: a 12-month managed operations subscription, per-active-agent commercial commitment, monthly SLA cadence covering agent behaviour drift monitoring, output quality measurement, escalation protocol for detected drift, and quarterly release cadence for agent configuration adjustments and portfolio composition decisions.
APPROACH AND TIMELINE
Approach and timeline
The subscription operates on a twelve-month cadence with monthly SLA reviews and quarterly release commitments.
Months 1-2 established baseline: comprehensive observability across all seven agents, drift monitoring instrumentation deployed per agent (model drift on the underlying LLM's response patterns, data drift on the input patterns each agent processes, behaviour drift on the agent's action patterns against defined operational envelope), output quality measurement calibrated per agent's business function.
Months 3-11 sustained operations: monthly SLA scorecards per agent covering drift indicators and output quality metrics, escalation protocol activation on any monitored metric crossing defined threshold, quarterly release cadence for agent configuration adjustments and portfolio composition decisions informed by operational evidence.
Month 12 executed portfolio review: full-year operational data reviewed against original per-agent SLA targets, retirement recommendations for agents where operational evidence supported it, onboarding recommendation for one new agent that had been in development during the year, subscription renewal scoping conversation.
Practice Lead attendance ran across all twelve monthly reviews. Escalation to CEO was invoked once, on the retirement recommendation for one agent where the business function owner had initially disagreed with the operational evidence — resolved with a joint review of the drift monitoring data and the alternative options.
WHAT WE RECOMMENDED AGAINST
What we recommended against
The recommendation the engagement was scoped to be willing to make.
Two elements the client's internal debate had assumed were in scope were recommended against during subscription scoping.
The proposed prompt engineering as recurring service — the CDO's team had partially assumed C9 would include ongoing prompt engineering work as part of the retainer — was recommended out of scope. Prompt engineering is build activity, not operations activity. Including it in the subscription would have blurred the boundary between "operations discipline that holds agent quality" and "build activity that changes agent behaviour", and compromised the drift monitoring signal — if the agent is continuously being modified, drift measurement loses meaning. Recommendation: prompt engineering handled as scoped change orders when operational evidence supported agent modification, not as included service in the recurring subscription.
The proposed immediate agent behaviour standardisation across the portfolio — a project to normalise how the seven agents responded to comparable inputs — was recommended against as an initial subscription scope element. The seven agents had been built by three implementation partners with different design conventions; standardisation would be significant work with uncertain business value, and would only be worth doing if the operational evidence surfaced actual business impact from inconsistent agent behaviour. Recommendation: monitor for six months to establish whether behaviour inconsistency actually creates business impact; revisit standardisation only if evidence supported the case.
Both recommendations were accepted. The subscription focused on drift monitoring and operational cadence, not on continuous agent modification. Behaviour inconsistency was monitored but not addressed pre-emptively; six months of operational data indicated only one agent pair generated meaningful user confusion from inconsistent behaviour, addressed via targeted change order rather than portfolio-wide standardisation.
OUTCOMES
Outcomes
Year-1 operational outcomes centred on the drift monitoring value. Three drift incidents were surfaced through monitoring before business impact materialised: one model drift incident on a customer service agent following the underlying LLM's vendor-side version update, one data drift incident on an operations support agent following a change in the input document format from an upstream system, one behaviour drift incident on a knowledge worker agent where output quality had gradually degraded through accumulated prompt modifications from the previous team. All three were resolved through change orders within acceptable timelines, with no business function reporting agent-related service degradation.
Two agents were recommended for retirement based on operational evidence: one had been superseded by native platform capability the underlying vendor released mid-year (agent was solving a problem the vendor now addressed at platform level), one had operational cost exceeding the business value the agent produced (the operational-value-per-cost analysis was conducted with the business function owner using the C9 subscription's operational data). One new agent was onboarded to production through the subscription's quarterly release cadence, with baseline drift monitoring in place from day one.
The CDO's stated Year-1 success measure — "agent portfolio managed with the same operational discipline as production applications" — was met.
WHAT COMES NEXT
What comes next
The subscription renewed into Year 2 at a scoping conversation between the CDO and Practice Lead in Month 11. Year-2 scope expands to include a second business function's agent portfolio (three additional agents built during Year 1), maintaining the per-active-agent commercial model. The behaviour standardisation question that the initial subscription recommended monitoring rather than addressing has been re-opened as a scoping conversation informed by the operational evidence.
Adjacent SKUs: C9 · B14 Agentic Workflow Agent Build™ · C8 MLOpsRun™
Agents in production and the internal team drowning?
Book a clinic. Practice Lead attends. Agent operations as monthly SLA cadence — not prompt engineering as reactive whack-a-mole against user complaints.