Esc back to picker · ← → nav · O overview · F fullscreen
Digital Teammates · For the Agents Team
From agents to digital teammates: a working hypothesis.
Press → to begin
Press O for slide overview
Digital Teammates · Working HypothesisFor internal discussion1
Executive SummarySection 1 of 5 · Context
Today's agents are deeply powerful — but a teammate is a different kind of thing, and across four dimensions the gap is concrete.
Dimension
An agent today
A teammate
Identity
Service account; acts "on behalf of" a user
First-class Entra ID; acts as itself, with its own audit trail
Proactivity
Waits to be summoned; passive between sessions
Engages on signal; follows up once, then backs off
Composability
Independent silo — users broker every handoff
Calls other teammates directly; orchestrates without the human
Receipts
Answers; sources optional, confidence implicit
Citations, confidence, named escalation by default
So whatThree prototype scenarios — Crisis Response, New Hire, Demand Spike — stress different combinations of these dimensions. Any single shift in isolation is weak; only the combination changes the product.
Digital Teammates · Executive SummaryFor internal discussion2
ContextWhere we are today
Each of our agents is good at its job — and each has a clear next chapter from "software you summon" to "teammate you work with."
How each agent is positioned today — and where the headroom is.
Agent
What it does today
How users invoke it
Where the headroom is
ESS
Room booking, drafting, HR self-service, IT ticket flow
@-mention in chat, profile card
Proactive engagement; continuity that follows the user across moments
Showing up between sessions; signal-driven coaching at the right moment
Workforce Insights
People recommendations, team signals, account insights
Direct queries from managers and leaders
Surfacing shifts proactively; org-level signals to leaders without being asked
The opportunityEach agent has built a strong foundation. The next chapter is in the connective tissue between them — identity, proactivity, composability, receipts — which is where agents become teammates.
Digital Teammates · ContextFor internal discussion3
HypothesisSection 2 of 5 · Framework
Moving from agent to teammate is a four-dimensional shift — not a single feature, and not a rebrand.
Each dimension is a design and engineering commitment. The framework is how we evaluate scenarios and pace investment.
01
Identity
Real Entra ID. Per-teammate audit. "Acting as," not "acting for."
Implies: every teammate is a directory entity with its own scoped permissions and lifecycle.
Cost: heaviest of the four — identity is platform work, not feature work.
02
Proactivity
Shows up on signal, not summon. Follows up once if you don't engage, then stops.
Implies: shared signal model and back-off semantics across agents — not per-team thresholds.
Cost: medium; depends on signal infrastructure already in flight.
03
Composability
Teammates call teammates. Cross-agent handoffs without human brokering.
Implies: contract for confidence, citations, and refusal that every agent speaks.
Cost: medium; needs an interop spec, no new identity work.
04
Receipts
Citations on claims, confidence on recommendations, named escalation when out of depth.
Implies: shared UX spec for "showing your work" — every agent renders it the same way.
Cost: smallest of the four — highest UX leverage per unit work.
NoteThese shifts are not internal jargon — they are how the industry describes the move from copilot to agent to "AI worker." We just need shared definitions across our agents.
Digital Teammates · FrameworkFor internal discussion4
Evidence PlanSection 3 of 5 · Scenarios
Three prototype scenarios stress different combinations of the four shifts — together they cover the full framework.
No single scenario "proves" the model. The cross-coverage is the point.
Scenario
Trigger
Identity
Proactivity
Composability
Receipts
Headline outcome
01 — Crisis Response
P1 incident; ESS in the squad chat
Deployment plan + readiness paths in minutes
02 — New Hire
HR record created; Day 1 ramp
Coordinated setup + manager readiness snapshot
03 — Demand Spike
47% spike in benefits queries
System-level adaptation, not just per-user reply
Primary stressSecondaryOut of scope this scenario
Reading the matrixEach scenario is a forcing function for two or three shifts. Where a shift shows up across multiple scenarios — Proactivity, Composability, Receipts — that's where investment compounds.
Digital Teammates · Scenario CoverageFor internal discussion5
A 47% spike in benefits questions: ESS treats it as a system-level signal, asks WFI for context, and adapts behavior — not just one answer.
Proactive. Notices the pattern before users surface it.
Composable. Asks WFI for org-level signal.
With receipts. Cited guidance; suggested comms.
Teammate flow
IdentityProactivityComposabilityReceipts
Signal
47% spike in benefits queries detected.
ESS
Frames as system-level pattern.
WFI
Returns context: open enrollment approaching.
ESS
Adapts user-facing answers with cited guidance.
ESS
Flags admins; proposes targeted comms.
Live prototype
Digital Teammates · Scenario 03For internal discussion8
SynthesisSection 3 of 5 · Cross-scenario read
The four shifts are not independent investments — each becomes useful, or risky, depending on the others.
Each row below is a way our roadmap can go wrong if we sequence wrong.
If we ship…
Without also…
The failure mode is…
Why
Proactivity
Identity
Surveillance
Unscoped reach-outs with no audit trail are indistinguishable from a black-box bot.
Composability
Receipts
A black box
When ESS calls AILA which calls WI, the user has to see the trail — otherwise the handoffs hide the evidence.
Identity
Proactivity
Just SSO
New directory entity, same passive software. The user notices nothing new.
Anything
Composability
Parallel ecosystems
Each team rebuilds learning, signals, policy primitives — instead of sharing them.
ImplicationInvestment sequencing matters: identity + receipts should land before — or alongside — proactivity and composability. Otherwise the user experience is worse than today.
Digital Teammates · SynthesisFor internal discussion9
Build ApproachSection 4 of 5 · How
From Word doc to working prototype in days, using a doc → Figma → CLI → live pipeline.
The pipeline itself is the second exhibit: it's the kind of thing we should be using to settle product debates.
Step 01
Written argument
Started in Word with the ESS-centered proposal — scenarios, principles, dependencies. Forced clarity before any pixels.
→
Step 02
Figma exploration
Translated scenarios into Teams surfaces: agent avatars, color system, walkthrough bar. Stopped arguing in prose.
→
Step 03
GitHub Copilot CLI / Claude Code
Used GitHub Copilot CLI or Claude Code in the terminal — wired to the Figma MCP server — to translate the design into a runnable HTML prototype. Iterated on copy, timing, and interaction with the model in the loop.
→
Step 04
Live prototype
Three scenarios with embedded walkthroughs and deep links. This deck embeds it directly — argument and artifact in one page.
So whatTwo weeks ago this was a Word doc. Today it's a clickable artifact. The cost of being concrete just collapsed — and that should change what kinds of pitches we make to each other.
Digital Teammates · Build ApproachFor internal discussion10
An observationSection 4 of 5 · How
If we can prototype future products in days, the shape of how we argue about them changes too.
A side-effect of the build pipeline we used here.
Before
After
Format
Doc → review → doc → review
Doc → prototype → critique
Disagreement happens
At the level of words and intentions
Against runnable, observable behavior
Time-to-decision
Weeks per round
Days per round
Cost burden
Reviewers fund the work ("build it for me first")
Originator fronts the cost, once — CLI + LLM
What gets argued
Whether the idea is right in principle
Where the prototype is wrong in practice
Digital Teammates · New way of workingFor internal discussion11
Open questionsSection 5 of 5 · Discussion
Four questions I'd like the team's input on — one per shift.
01
Identity
What does a teammate-shaped Entra identity look like?
Scope: per-teammate permissions; "acting as" vs. "acting for" semantics.
Lifecycle: provisioning, refresh, deactivation — owned by which team?
Audit: minimum-viable accountability trail per teammate action.
02
Proactivity
What's the shared back-off model across our agents?
Threshold: what signal strength justifies first engagement?
Cadence: follow up once and stop, or staged escalation?
Fragmentation risk: if each agent picks its own rule, the experience splinters.
03
Composability
What's the inter-teammate contract?
Schema: confidence, citations, refusal — what every call returns.
Errors: who escalates when, to whom, with what context attached?
04
Receipts
How should a teammate show its work?
Citations: what's a sensible format for sources?
Confidence: when does a teammate say how sure it is?
Refusal: what does a teammate say when it shouldn't answer?
Digital Teammates · Open questionsFor internal discussion12