Agentic AI Capability

Ten task agent roles: one runs in production, nine are designed.

Juno’s agent blueprint is a design of 960 agent roles across five end-to-end chains: the map of what Juno can be configured to do, built one role at a time, the way the PO cancellation agent was.

Cover and stockwhat to hold, and whereSafety stockA2Excess stockA2Stock rebalanceA2AllocationA2Purchase order flowwhat to bring in, and whenPO cancellationA2ExpediteA2De-expediteA2Shortage triageA1Supply performancewho delivers, and how wellLead time driftA1Supplier on-time deliveryA1Ten task agent rolesone in production · ninedesigned, yet to be builtblueprint · designed, yet to be built · PO cancellation runs in production · the level beside each role is its designed start
Three groups of roles, one engine underneath. The chip beside each role is its designed starting level; the PO cancellation agent is the one in production. Blueprint: designed, yet to be built.
1
task agent in production: PO cancellation, at a high-tech manufacturer
production
9
task agent roles designed in the blueprint, yet to be built from the same parts
blueprint · designed, yet to be built
A1 · A2
designed starting levels, raised agent by agent as evidence arrives once a role is built
blueprint
212
approvals in six weeks before the PO cancellation agent moved up a level
showcase estate
Where this sits

Stage 5 of six: act

Step by step

What every task agent does, from trigger to action

Five steps, the same for every role, so oversight is learned once.

WHAT EVERY TASK AGENT DOES, IN FIVE STEPS01A trigger firesa schedule, an interval,an event in a source, or aperson02Facts are gatheredmapped data read acrossyour systems, where itlives03Rules areevaluatedthe published ruleversion, applied the sameway each run04A finding isrecordedwith a recommendation, itsbasis and a confidence05The agent acts atits levelinform, draft for a namedapprover, or executeinside limitsevery accept, edit and override returns to the agent as evidence for the next run
The same five steps run for the PO cancellation agent today and for every role built after it, so one audit record covers them all.

An agent starts when something happens: the clock reaches a set time, an interval elapses, a row changes in a source, or a person runs it. It then gathers the facts it needs, reading the mapped tables through the federated query engine, so the rows stay in the systems that own them.

The Lima Rules Engine evaluates the published rules against those facts. The same facts and the same rule version give the same result on every run.

Where a rule matches, the agent records a finding. Where the evidence supports it, the finding carries a recommendation with a basis, a confidence and the value at stake.

What happens next follows the level you set. At A1 the agent informs a person; at A2 a named person approves its draft; at A3 it executes inside your limits and routes exceptions to a person. Every path ends in one record: rule, version, facts, decision and outcome.

The ten

What each role watches, works out and does, and which one runs

Each card names the signal, the calculation and the action a role is designed for, so a reviewer can judge scope in one reading.

PO cancellation

A2

In production. Cancels purchase order lines that stand above the need, inside their cancellation window, under a named approver.

Watches
Open lines, the demand behind them, each line’s cancellation window.
Works out
Projected excess in weeks of demand, the value at stake, days left.
Does
Drafts the cancellation; at A3 it cancels lines under your value limit.

Safety stock

A2

Designed role. It would keep the cover on each material in step with how demand and lead times behave.

Watches
Demand variability, lead time variability, the service level per material.
Works out
The cover that holds that service level at today’s variability.
Does
Proposes a new safety stock value for the planner to approve.

Expedite

A2

Designed role. It would pull supply in where a requirement falls short and the supplier can hold the date.

Watches
Requirements that fall short, open supply, confirmed supplier dates.
Works out
Which pull-in closes the shortage, what it costs, how firm the date is.
Does
Drafts a date change on the line for the buyer to approve.

De-expedite

A2

Designed role. It would push supply out where it lands well ahead of the demand behind it.

Watches
Incoming supply, the demand behind it, the cover floor.
Works out
How many weeks the line can move out while cover holds.
Does
Drafts a later date, with the working capital it frees beside it.

Shortage triage

A1

Designed role. It would turn a morning of exception messages into a ranked list a planner works through.

Watches
Exception messages, open orders, the stock behind each shortage.
Works out
The cause of each shortage, the orders at risk, the order to work in.
Does
Hands over the ranked list with the evidence for every line.

Excess stock

A2

Designed role. It would find stock and open supply that run ahead of forward demand.

Watches
Stock on hand, stock on order, forward demand per plant.
Works out
Weeks of cover, the value above policy, where the part runs short.
Does
Proposes a move, a hold on further supply, or a finance review.

Lead time drift

A1

Designed role. It would report where a planned delivery time has drifted away from real supplier behaviour.

Watches
Planned delivery time against the dates suppliers deliver on.
Works out
The drift per supplier and material, and its effect on planning.
Does
Reports the drift and proposes a new planned delivery time.

Supplier on-time delivery

A1

Designed role. It would keep one scorecard of supplier delivery performance across every system.

Watches
Confirmed dates, receipt dates and quantities, by supplier and site.
Works out
On-time in-full by supplier, material group and period, with the trend.
Does
Publishes the scorecard and flags suppliers below the agreed line.

Stock rebalance

A2

Designed role. It would cover a shortage at one site from surplus at another, when it earns its cost.

Watches
Stock and demand at every plant and warehouse.
Works out
Which transfer closes a gap, its freight cost and its lead time.
Does
Drafts the transfer for approval, with cost and timing shown.

Allocation

A2

Designed role. It would split constrained supply across customers and channels by your policy.

Watches
Constrained supply, open demand, the priority each customer carries.
Works out
A split of the available quantity that follows your policy.
Does
Drafts the split; the confirmed result goes back to the order.

The PO cancellation card is the agent in production. The other nine cards are roles designed in the agent blueprint, yet to be built; each states what the role would watch, work out and do once it is built from the same parts. productionblueprint

Levels

Where the blueprint starts each role, and what moves it up

The level travels with the agent and its version history, so an auditor sees what was allowed and when.

0A0 · Human-ledThe work stays with people;the blueprint lists the stepso the map stays complete.1A1 · AssistedThe agent informs; the humandecides and acts.2A2 · SupervisedThe agent decides and drafts;a named human approves eachaction.3A3 · DelegatedThe agent decides and executeswithin versioned rules; thehuman handles exceptions only.4A4 · AutonomousThe agent executes andself-corrects; the humanaudits after the fact.product · the level is a setting on each agent, held in the rule store with its version history
A2 supervised is the designed starting point for most roles: the agent drafts, and a named person approves each action. The PO cancellation agent started there.

The blueprint designs eight of the ten roles to start at A2 supervised, and shortage triage, lead time drift and supplier on-time delivery to start at A1 assisted, where the agent reports and the person acts. These are designed starting levels. The level is a setting on the agent, and it is yours to move once a role is built.

Raising a level takes three things: a passing test on recorded cases, the named owner’s approval, and a live escalation route. Lowering a level takes one step and applies at once.

Market practice supports the cautious start. Capgemini found trust in fully autonomous agents fell from 43% to 27% in a year (Capgemini, 16 July 2025), and a KPMG pulse survey reports 61% of organisations requiring approval processes for agent actions (KPMG, Q2 2026).

The autonomy ladder explains all five levels, and the use cases chapter shows the whole blueprint of 960 designed roles in the same frame. blueprint

The running example

Monday 07:40: the one agent that runs, and the roles that would join it

Showcase estate. The shortage triage and excess stock roles are designed, yet to be built.

The running example

L. Tan opens 1,240 exception messages; 212 of them touch an open purchase order line. The PO cancellation agent, the one role on this page that runs, takes those lines, applies the cancellation policy, and finds 38 lines worth $412,000 above the need with a window that closes on Thursday.

D. Okafor approves the drafts in the queue. After six weeks and 212 drafts, 203 approved as written and 9 edited, lines under $25,000 moved to A3 and D. Okafor sees exceptions only on those. A shortage triage role, once built from the blueprint, would rank the morning’s messages before she opened them; an excess stock role would confirm that the same materials carry more than eight weeks of cover.

Next capability

From intent to a reviewable plan

How a policy stated in plain language becomes a planning table your team reviews, edits, simulates and publishes.

Terms on this page
Task agentA named agent for one recurring supply chain decision, with its own rules, limits and level.
Agent blueprintLima’s design of 960 agent roles across five chains: the map of what Juno can be configured to do, built one role at a time.
TriggerWhat starts an agent run: a schedule, an interval, an event in a source, or a person.
FindingA recorded match between a published rule and the facts at that moment.
Designed starting levelThe level the blueprint proposes for a role before it is built; your setting replaces it.
Escalation routeThe named desk an exception or an overdue approval goes to next.
Questions
Can we change what a role does?

Yes. A role’s rules are yours to write and edit in plain language and in the rule builder. The blueprint’s design is the starting point, and every edit becomes a new version.

Can we add roles of our own?

Yes, the same way: state the intent, review the plan the Planner Agent drafts, simulate it and publish it. The agent blueprint holds 960 designed roles to start from, and a role of your own is built from the same parts.

Do all ten have to be built together?

They are built one at a time. Start with the one that saves money on open orders, which is the agent that runs today, and add the role that ranks the morning work next; the rest follow at your pace.

What happens when two agents disagree?

Arbitration ranks the competing recommendations inside a band you set; beyond the band a named person decides. The arbitration page works a case through.

In the platform

Where Juno sits among the four capabilities

Each one builds on the one before it.

One shared foundationCatalogLineageAccess policyWrite path

See Juno answer and act on your own data

Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.