Microsoft is turning workforce planning into an agent workflow. People Skills can infer work-related skills from Microsoft 365 activity and profile data. Workforce Insights Agent can help leaders understand team composition, skills gaps, hiring pressure and upskilling priorities from organisational data, custom attributes and People Skills data.

That gives leaders a faster way to ask workforce questions. It also creates a sharper operating problem: skills data is only useful when the company knows where it came from, who confirmed it, which attributes are authorised, what the agent is allowed to show, and which decision the insight is meant to support.

Workforce insights agents need skills memory. Skills memory is the governed record of inferred skills, confirmed skills, HR attributes, employee controls, manager context, source ownership and decision outcomes that keeps AI workforce planning from becoming another confident dashboard with weak operating evidence.

Workforce planning is becoming an agent surface

Microsoft’s Workforce Insights Agent documentation frames the agent around organisational composition, workforce planning, skills gaps, organisational structure and filtered team search by attributes such as level, location and title. Microsoft also states that responses are based on the user’s role, admin policies and available team information.

The data model matters. The agent uses Organizational Data in Microsoft 365, custom attributes sent to Workforce Insights and People Skills data. Microsoft’s organisational-data guidance says Workforce Insights relies on clean, governed HR data to provide a trusted view of organisational relationships and composition. People Skills adds the skills layer, with AI-generated personalised skill profiles mapped to a skills taxonomy.

The Microsoft 365 Champion recap makes the management use case clearer. Leaders can query organisational data conversationally, see real-time insights and visualisations, identify hiring or skills gaps, and guide talent movement or upskilling. Microsoft also notes the agent is in the Frontier programme, with no confirmed general availability timeline in that recap.

That is a serious shift. Workforce planning is moving from analyst reports and HRIS screens into conversational decision support. The pressure point is data interpretation, not access to a prettier interface.

Inferred skills are evidence, not judgement

People Skills can generate personalised skill profiles from Microsoft 365 Graph signals, profile data, role and work activity. Users can confirm, edit or remove skills, add their own skills manually, opt out of inferencing and control sharing. Microsoft’s responsible AI guidance for People Skills also excludes behavioural monitoring, performance management and employment decisions from the service’s intended use.

That boundary deserves respect. A skill inferred from documents, meetings and collaboration patterns is a clue about work exposure, with limited authority over capability, depth, readiness or promotion risk.

Skills memory gives that clue context. It records whether the skill was inferred or confirmed, when the profile changed, which taxonomy label was used, whether the employee has shared it, whether the manager has reviewed it, and which decision is allowed to use it. Without that record, an agent can make a staffing conversation feel more precise than the data deserves.

That matters when money and careers sit behind the question. A leader asking for AI skills across a team needs to know whether the answer reflects confirmed capability, inferred activity, imported HR data, outdated titles or an incomplete taxonomy. The agent can summarise the pattern. The organisation still needs an operating memory that explains the confidence level behind the pattern.

HR attributes need source authority

Organisational data looks simple until the agent starts counting managers, levels, locations and team structures. Microsoft documents a SupervisorIndicator column that can override the default manager logic for Workforce Insights chat responses and analytics. That column becomes the authoritative signal for manager status, while reporting hierarchy, access control, delegation and privacy enforcement still rely on direct reporting relationships.

That distinction is exactly why skills memory belongs in the operating layer.

A company can have formal managers, matrix leads, contractor sponsors, project owners and regional heads who all carry different kinds of authority. The agent needs to know which definition controls each question. Counting people managers for span analysis is different from finding the right sponsor for an upskilling plan. Filtering by job title is different from deciding who can approve a role change.

Skills memory should preserve the source hierarchy for workforce attributes:

  • which HR system owns the attribute
  • which fields are public, confidential or custom
  • which app can access each attribute
  • which manager definition applies to the insight
  • which delegate can view the report
  • which employee controls affect skills visibility
  • which disclaimer or internal policy frames interpretation

Those records stop a workforce insight from drifting into unreviewed authority. The agent can answer faster, but the company still decides which data is controlling evidence.

Workforce recommendations need decision residue

A useful workforce agent does more than report the current team shape. It pushes leaders towards decisions: where to hire, which skills to develop, who can cover a capability gap, how a team mix compares with another group, and where AI adoption changes the work.

Microsoft’s 2026 Work Trend Index argues that leaders need to rearchitect work as agents take on more execution. It also reports that organisational factors account for twice the reported AI impact of individual effort alone, based on Microsoft’s cited survey and productivity-signal analysis. That supports the operating-layer thesis: individual AI use matters, but the company captures value when workflow, management support, standards and talent practices change around the work.

For workforce agents, the learning loop should not end at the answer. If the agent identifies a skills gap, the company needs to record what happened next. Did the leader accept the recommendation? Did HR challenge the data? Did the manager know the employee had unrecorded experience? Did a hiring plan change? Did the organisation create a learning path, move talent, or update the skills taxonomy?

That residue becomes skills memory. It turns one workforce query into reusable operating knowledge. The next leader can see which skills signal proved useful, which profile needed correction, which role definition created confusion and which decision created measurable improvement.

This connects directly to AI onboarding agents needing ramp memory, AI analytics agents needing metric memory and Copilot Studio agents needing evaluation memory. Workforce data, onboarding plans, metrics and evaluations all fail in the same place when the organisation does not preserve the correction path.

Build the first skills loop around one decision

Teams do not need a perfect enterprise skills ontology before using workforce insights. They need one decision with enough pressure to expose whether the data, review path and privacy boundary hold.

Start with a specific question, such as staffing an AI initiative, identifying upskilling priorities for a product team, finding internal experts for a customer problem, or reviewing span and layer pressure after a reorganisation.

Then map the loop:

  • what the leader is allowed to ask
  • which organisational data and custom attributes are in scope
  • which skills are inferred, confirmed or imported
  • which employee controls and admin policies apply
  • which HR or business owner reviews the answer
  • what evidence supports the recommendation
  • where corrections to profiles, taxonomy or org data go
  • how the final decision is recorded for future workforce planning

That loop is the difference between an AI workforce dashboard and governed company memory. The dashboard answers a question. Skills memory improves the next decision.

Where Model Operator fits

Model Operator builds the memory and workflow layer teams need before AI can be trusted cross-functionally. For workforce insights agents, that means mapping source authority, permission boundaries, review paths, decision logs and write-back routes around the data that leaders already use to make people and operating decisions.

The package route depends on the first constraint. Fragmented skills and organisational context points towards an Agentic Company Brain. AI questions inside Slack or Teams point towards Company Brain + Slack / Teams Bots. Leadership uncertainty around where agents belong points towards AI Initiative Consulting.

If your team is bringing workforce insights, company knowledge or internal AI agents into Microsoft 365, start with one pressured decision. Map the data, owner, review path and correction loop before widening access.

Model Operator builds governed company memory and internal AI interfaces for teams that need AI to work inside real operating constraints.

Start a build conversation: modeloperator.io or alexander@modeloperator.io.