BLUEPRINT · FAST STAGE 3 · SYSTEM-LEVEL AGENTIC CAPABILITIES
The Twenty Agentic Capabilities
More than an autonomous agent. For years your function rationed its best thinking to the cases that could bear the cost — everyone else got the template. That was never a choice; it was the only way it could work. It isn't any more. Each of the twenty below is one of those compromises, now optional.
All twenty are buildable today on Copilot Studio, Dataverse and the Power Platform. This is system-level Digital Labour — not desktop AI.
Humans on judgment. Agents on the rest.
Get your free exposure reportHow to read the tags
Every capability carries its move — the work it takes, or the call it sharpens.
REPLACE
Work taken off your people
The agent carries it — bounded, audited and answerable.
ENHANCE
A call made sharper
The agent reasons in depth; your person decides better.
HUMAN GATE
Where authority lives
Two capabilities — the deliberate keep-points where a person says yes, and where a person says stop.
These twenty aren't a product list — they're the frame. Each is a reusable pattern: what it does never changes; what it's applied to is defined per function, per wave. The capabilities you haven't scoped yet are already named, already layered, already governed.
Decisioning
How the system reasons — and why the call gets sharper.
01
Deliberative Council
Today, most cases get one opinion — not a panel.
ENHANCE
Explore →
Deliberates; it never decides.
02
Constraint Hierarchy
This is the floor nothing drops below.
ENHANCE
Explore →
A breach is blocked outright — never weighed against cost.
03
Surfaced Dissent
Today, the expert who disagreed gets averaged away.
ENHANCE
Explore →
Records dissent; never resolves it for the human.
04
Optioned Recommendation
Today, you get one answer to rubber-stamp.
ENHANCE
Explore →
Frames the choice; the human makes it.
Orchestration
How the work is planned, shaped and gathered for each case.
05
Generated Workflow
Today, every case gets the same template.
REPLACE ENHANCE
Explore →
Routine tasks run alone; high-stakes ones hold for a person.
06
Per-Case Personalisation
Today, everyone gets the form letter.
ENHANCE
Explore →
Anything affecting a person's outcome is human-approved.
07
Adaptive Intake
Today, people re-key what a form already asked.
REPLACE
Explore →
Brings in a human the moment a risk signal appears.
Assurance & the Human Gate
The record, the watch — and the one place a person says yes.
08
Governance by Design
This is what makes the rest safe to run.
REPLACE
Explore →
Nothing happens off the record.
09 · THE HUMAN KEEP-POINT
HITL Control Points
This is where your authority stays.
HUMAN GATE
Explore →
Nothing consequential is written or sent without a human's yes.
10
Supervision & Escalation
Today, a case only gets looked at when someone checks.
REPLACE
Explore →
Escalates to a person — never acts around one.
Continuity & Learning
Judgment that carries through the case, and compounds after it.
11
Lifecycle Continuity
Today, the thread is lost at every handover.
ENHANCE
Explore →
Continuity of judgment; authority stays with your people.
12
Outcome Learning Loop
Today, every case starts from zero.
ENHANCE
Explore →
Proposes learnings; never silently rewrites its own logic.
Foundation
The supporting services everything above stands on.
13
Grounded Retrieval
Today, the answer rests on what someone remembered.
REPLACE
Explore →
Informs the reasoning; it doesn't decide.
14
Memory
Today, the case asks you to repeat yourself.
REPLACE
Explore →
Recalls and informs; it never acts on its own.
15
Document Extraction
Right now, someone is re-keying paperwork by hand.
REPLACE
Explore →
Low-confidence reads route to a person, never committed blind.
16
Validation
Today, bad inputs are caught downstream — expensively.
REPLACE
Explore →
Failures are held and routed back — never pushed through.
Action
Where a decision becomes a done thing — under a person's OK.
17
Proactive Initiation
Today, a case waits in a queue for someone to notice.
REPLACE
Explore →
Low-stakes nudges fire alone; commitments wait for approval.
18
Action / Tool Execution
Today, the doing waits for a free pair of hands.
REPLACE
Explore →
Every consequential action passes a human gate before it commits.
AI Command
The name and the stop — every agent identified, any agent stoppable.
19
Agent Identity
Today, no one can say how many agents are running — or whose they are.
REPLACE
Explore →
Every agent has a name, a purpose and an accountable owner.
20
Shutdown Authority
Today, stopping a system means finding the right person — then a meeting.
HUMAN GATE
Explore →
A named owner can stop any agent in one pre-authorised move.
The governed record beneath all twenty
Agents reason. The record holds. Every capability reads from, and writes to, one governed data foundation inside your own Microsoft environment.
Fixed classifications
Agents choose from locked lists, such as stage, severity and pathway, never from free text. Every output stays consistent, comparable and reportable.
A record that can't be rewritten
Every recommendation, decision and hold lands in an append-only log. What each agent saw, and what each person decided, stays on file.
Change behind a human gate
Thresholds and lists are configured, not coded. Changing them is a human decision at a gate, because it changes every case that follows.
One governed function — built capability by capability
The routine is carried by agents that are bounded, audited and answerable. The judgment stays with your people. These twenty are the Blueprint stage — start with a free read of your function, then talk it through with us.
Humans on judgment. Agents on the rest — the thinking unrationed, the grind lifted, the record kept, the unprompted work done.
Get your free exposure report