Microsoft Work IQ API for Persistent Agent Context and Memory

Microsoft Work IQ is a production-ready intelligence layer that gives AI agents persistent, durable memory across sessions and applications. Available as a set of APIs within Microsoft 365 Copilot extensibility, Work IQ enables agents to read, write, and reason over work context — learning how users work, what they care about, and what needs to happen next. The agentic reasoning step occurs when Work IQ evaluates the current work context against historical patterns and determines what information from past sessions is relevant to the current task — it doesn't just retrieve all history, it selectively surfaces the context that matters. This is agentic because the memory layer makes decisions about relevance, not just storage. Microsoft Scout is the first agent built on Work IQ, but the APIs are available for any agent to use.

BUSINESS PROBLEM

Every AI agent today suffers from amnesia. A customer support agent that solved a complex issue last week has no memory of the solution when the same customer calls again. A coding agent that learned your project conventions yesterday forgets them today. According to Microsoft's 2026 agent survey, 78% of developers say 'lack of persistent memory' is the primary reason they cannot trust AI agents for long-running or recurring tasks. The challenge is not just storage — it's relevance. Storing everything creates noise. Storing nothing creates amnesia. Work IQ solves the relevance problem by building a model of what matters.

WHO BENEFITS

Enterprise developers building custom AI agents: your agents currently start fresh every session, frustrating users who must re-explain context. Work IQ APIs provide turnkey persistent memory — your agent remembers user preferences, past interactions, and project context across sessions. ISVs building AI-powered SaaS applications: your users expect your AI features to remember them across logins, devices, and sessions. Work IQ provides production-grade memory infrastructure without building it yourself. Enterprise IT teams deploying Microsoft 365 Copilot: your users want Copilot to understand their work patterns, projects, and priorities. Work IQ is the infrastructure that makes Copilot contextually aware.

HOW IT WORKS

1. Context Ingestion: The agent sends work context to Work IQ via API — current project, active tasks, recent decisions, user preferences, and interaction history. Work IQ processes and indexes this data, building a structured model of the work context.
2. Relevance Scoring: When the agent needs memory for a new task, it sends a query to Work IQ. The API evaluates the query against stored context and scores each memory item for relevance across 3 dimensions: temporal recency, topical similarity, and task相关性 (correlation based on user's work patterns). Low-relevance memories are filtered out.
3. Context Retrieval: Only the highest-scoring context items are returned to the agent. The result is a focused context package — typically 3-7 items — that provides exactly what the agent needs without overwhelming its context window.
4. Memory Write: After completing a task, the agent writes a new memory entry — what was accomplished, what decisions were made, what context was important. Work IQ indexes this for future retrieval.
5. Cross-Session Continuity: When the same user returns in a new session — possibly on a different device or in a different app — Work IQ retrieves the relevant context from the user's history. The agent picks up where it left off.
6. Learning Loop: Over time, Work IQ builds a model of the user's work patterns — who they collaborate with most, what times they prefer for deep work, which projects are active priorities. This model makes memory retrieval increasingly relevant over time.

TOOL INTEGRATION

Work IQ APIs (Microsoft, June 2026): Production-ready memory APIs for AI agents. Available through Microsoft 365 Copilot extensibility and Microsoft Graph. Currently in preview. Gotcha: Work IQ APIs require Microsoft 365 E3 or E5 license. Not available for standalone use without Microsoft 365 subscription.

Microsoft 365 Copilot (Microsoft): The primary surface for Work IQ-enabled agents. Provides the runtime environment, identity, and compliance controls. Requires Copilot license ($30/user/month). Gotcha: Copilot's agent runtime has a 5-minute timeout for synchronous agent tasks — long-running agents need async patterns.

Microsoft Graph API (Microsoft): The data source that feeds Work IQ. Provides access to emails, calendar, documents, chats, and contacts. Permissions managed through Entra ID. Gotcha: Graph API permissions require admin consent for many data types. Plan for a 2-3 week permissions review process with your IT security team.

ROI METRICS

1. Cross-session agent accuracy: 40% accuracy without memory → 85%+ with Work IQ context retrieval (Source: Microsoft Work IQ Technical Preview, 2026)
2. User re-explanation time: 5-10 min per session → 0-1 min with persistent context
3. Agent task completion rate: 55-65% without context → 80-90% with Work IQ-retrieved relevant history
4. Development time for memory infrastructure: 4-8 months building custom memory → 1-2 days integrating Work IQ APIs
5. Time to first ROI: measurable in the first week — first cross-session context recall saves a user 5+ minutes of re-explanation

CAVEATS

1. Work IQ APIs require Microsoft 365 E3/E5 licensing and Copilot subscriptions. Not suitable for organizations not already on Microsoft 365.
2. The APIs are in preview as of June 2026. Pricing, rate limits, and data retention policies are subject to change. Microsoft has not committed to GA pricing.
3. Memory relevance depends on data quality. If users routinely work across unconnected projects without clear context signals, Work IQ's relevance scoring degrades.
4. Cross-session memory raises privacy concerns. Organizations must implement clear data retention policies and user controls for memory data. Work IQ provides data lifecycle APIs but does not enforce retention — that's the organization's responsibility.