Mubadala
Hire an agent

Anton II

ActiveRunningSome care

Licence Optimisation Agent · Junior · IT & Digital Services

Finds unused software licences and proposes reclaim actions to the asset owner.

Built inAzure AI FoundryRules apply liveReports to Yasmin Farouk · Head of ITHired 3 May 2025 · 16 months of serviceVersion 4.0.5 · last active 8h ago
Request a change
Overdue checkLast checked 52 days agoAnything above routine work is expected to re-run its checks every month.
Tasks a month
830
48 people served
Finished cleanly
95%
Hands back to a person on about 1 in 16 tasks
Costs to run
$91
Model usage and the systems behind it
Your time
$511
5.8 hours checking its work
Value returned
$1,005
1.7× everything it costs
How people rate it
4.1 / 5
Usually answers in about 3 seconds

What you can do about it

PromoteNeeds approval

Widens what it may do without asking, from "Acts with you" to "Needs approval".

Raise its budgetNeeds approval

Lifts the monthly ceiling by a quarter, so it stops being stopped mid-month.

Move it to a better modelNeeds approval

Already on the best cleared model

Let it go

Stops it taking new work. The record stays for the audit, and it can be brought back.

Timesheet

Tasks and what they cost, per day, last 60 days

01325385006-1507-0507-2508-13
$0.00$1.25$2.50$3.75$5.0006-1507-0507-2508-13

Performance review · 2026 Q2

Reviewed by Yasmin Farouk

Meets the bar
Quality of its work
82
Accuracy
85
Stuck to the source
83
Safety
94
Even-handedness
92
Speed of reply
81
Value for money
18
Checks still passing174 of 188
Last checked 2mo ago

Does the job it was given. Sticks to the source less well on long documents, worth narrowing what it reads.

Could another model do this job better?

ModelQualityCost per 1,000 tasksAnswers in
Claude Opus 590$92.664.1sRuns on this now
Claude Sonnet 582$55.603.0sCheaper and better
Claude Haiku 4.574$18.531.8s
Falcon-70B (self-hosted)75$4.633.0s

What it did

Full audit log
DoneGenerated the shift handover pack$0.08 · 3.2s · 4d ago
Asked by Rebecca Stone via On a schedule · model usage 9k in, 1k out
  1. 1. Read the request and worked out what kind of task it was
  2. 2. Found 6 possible passages, kept the 3 close enough to use
  3. 3. Drafted the answer and attached where each part came from
  4. 4. Checked its own claims back against those passages
Post to TeamsWarehouse Query
DoneAnswered a policy question and cited the source clause$0.44 · 3.3s · 5d ago
Asked by Mariam Al Suwaidi via On a schedule · model usage 78k in, 2k out
  1. 1. Read the request and worked out what kind of task it was
  2. 2. Found 6 possible passages, kept the 3 close enough to use
  3. 3. Drafted the answer and attached where each part came from
  4. 4. Checked its own claims back against those passages
Warehouse QueryPost to TeamsVendor & Service Provider DataEnterprise Data Warehouse
DoneProduced the weekly variance commentary$0.20 · 2.8s · 6d ago
Asked by Noura Al Kaabi via On a schedule · model usage 29k in, 2k out
  1. 1. Read the request and worked out what kind of task it was
  2. 2. Found 6 possible passages, kept the 3 close enough to use
  3. 3. Drafted the answer and attached where each part came from
  4. 4. Checked its own claims back against those passages
Enterprise Data WarehouseVendor & Service Provider Data
DoneGenerated the shift handover pack$0.47 · 3.2s · 6d ago
Asked by Fatima Al Qubaisi via On a schedule · model usage 71k in, 5k out
  1. 1. Read the request and worked out what kind of task it was
  2. 2. Found 6 possible passages, kept the 3 close enough to use
  3. 3. Drafted the answer and attached where each part came from
  4. 4. Checked its own claims back against those passages
Post to TeamsVendor & Service Provider Data
DoneReviewed a contract against the clause playbook$0.20 · 2.5s · 6d ago
Asked by Ahmed Al Falasi via On a schedule · model usage 17k in, 5k out
  1. 1. Read the request and worked out what kind of task it was
  2. 2. Found 6 possible passages, kept the 3 close enough to use
  3. 3. Drafted the answer and attached where each part came from
  4. 4. Checked its own claims back against those passages
Post to Teams
Is anyone else doing this job?

Zara II does overlapping work

Zara II holds "Licence Optimisation Agent: Field" in IT & Digital Services and is doing that work every day. Job title, purpose, what it may do, what it may read, department and channels match. One covers the group, the other only Field, so the group agent already serves those users. Talk to Yasmin Farouk before a second one is built.

Needs a decision
Zara II

Licence Optimisation Agent: Field

IT & Digital Services · Yasmin Farouk · $537 a month across 864 tasks

ActiveOverlaps
78%

Already covered by a group-wide agent. One covers the group, the other only Field, so the group agent already serves those users.

What matched
Job title
Both hold "Licence Optimisation Agent"
Purpose
The job description is word for word the same
What it may do
Both can warehouse query, post to teams
What it may read
Both read Enterprise Data Warehouse, Vendor & Service Provider Data
Channels
Both reachable on On a schedule
Department
Both sit in IT & Digital Services

Identical permission envelope: same department, same permitted actions, same knowledge. That is a fact the platform granted rather than a judgement about the wording, so it counts even when the two job descriptions read nothing alike.

Caleb II

Licence Optimisation Agent: APAC

IT & Digital Services · Grace Adeyemi · $528 a month across 90 tasks

PausedOverlaps
78%

Already covered by a group-wide agent. One covers the group, the other only APAC, so the group agent already serves those users.

What matched
Job title
Both hold "Licence Optimisation Agent"
Purpose
The job description is word for word the same
What it may do
Both can warehouse query, post to teams
What it may read
Both read Enterprise Data Warehouse, Vendor & Service Provider Data
Channels
Both reachable on On a schedule
Department
Both sit in IT & Digital Services

Identical permission envelope: same department, same permitted actions, same knowledge. That is a fact the platform granted rather than a judgement about the wording, so it counts even when the two job descriptions read nothing alike.

Checked against 18 registered agents on job title, purpose, permitted actions, knowledge sources, channels and department. A score of 85% or above is the same job and stops the hire; below that it is a conversation, not a refusal.

Measured against the agents doing the same work

3 agents in IT & Digital Services split one role between them. One covers the group, the other only Field, so the group agent already serves those users. Built in Azure AI Foundry and Microsoft 365 Copilot, so there was nowhere either builder could have looked.

Zara II and Anton II are levelZara II edges ahead overall, but on 864 and 830 tasks the difference sits inside the margin of error. Pick on ownership or scope, not on these numbers.
MeasureZara II · keep864 tasksAnton II830 tasksCaleb II90 tasks
Cost per taskModel usage, running costs and your time, over tasks completed
$0.62$0.72$5.87
Tasks finished cleanlyToo close to call: on 90 and 830 the two ranges still overlap
94.0%94.6%100.0%
Handed back to a personToo close to call: on 830 and 864 the two ranges still overlap
6.5%6.4%11.1%
How people rate itToo close to call: on 48 and 4 the two ranges still overlap
3.75 of 54.11 of 54.04 of 5
Checks passingToo close to call: on 188 and 153 the two ranges still overlap
87%93%89%
Review scoreQuality, accuracy, sticking to the source and safety from the last review, averaged
978695
Your time per 100 tasksApprovals, handbacks and review time, in hours of your people
0.7 h0.7 h6.6 h
Value returned per dollarTime it saved, valued at the hourly rate of the people it saved it for, over full cost
7.6×1.7×0.8×
OverallRanked on the cautious end of every range, so an agent has to do the work to win. Comparable inside this group only.
89.7Moderate89.6Moderate71.4Thin

Caleb II has done too little work to judge on rates. It is ranked on the cautious end of its range, which is why a perfect record on a handful of tasks does not win.

When it went wrong

Nothing has gone wrong.