Memory and Retrieval Problems
| Worker Use Case | Customer Service Case Resolution |
| What it does | A customer contacts support with a problem. The AI worker reads the case thread, searches for similar previously resolved cases, retrieves the relevant policy or product documentation, identifies the most likely resolution path, and either provides a direct answer to the customer or drafts a response for the support agent to review. It also prepares a case summary capturing the issue, what was tried, and what was resolved, so any handoff to another agent or team is immediate rather than requiring the new person to re-read the entire thread. |
| What triggers it | A new support ticket, a live chat session, or a phone call transcribed into a case management system. |
| What it produces | A recommended resolution with source citations, a drafted response to the customer, and a structured case summary for the record. |
| Where the human sits | The support agent reviews the recommended resolution before responding, handles cases the AI cannot confidently resolve, and closes the case once the customer confirms resolution. |
| Where it shows up | Customer service centres across retail, financial services, SaaS, and telecoms, anywhere ticket volume outpaces human reading speed. |
| Why it matters | This card shows how Document Intelligence and Dialogue and Interaction combine in practice. Reading prior cases is document intelligence. Responding to the customer is dialogue. Most real workflows combine capabilities rather than exercising only one, and this card is one of the clearest demonstrations of that principle across the entire library. |
Examples
- Similar case finder — searches the case history for previously resolved issues matching the current customer’s problem and surfaces the resolution steps used, giving the agent a proven path rather than starting from scratch
- Case summary and handoff worker — reads the full case thread, produces a structured summary of the issue, actions taken, and current status, and prepares the receiving agent with everything they need before they engage the customer
- Customer issue resolution assistant — retrieves the relevant product documentation, policy, or knowledge article, drafts a resolution response for the agent to review, and suggests whether the case can be closed or requires escalation
- Refund processing agent — reads the customer’s refund request, checks it against purchase records and refund policy parameters, recommends approval or rejection with a policy citation, and routes exceptions to a supervisor
- Transaction dispute handling assistant — retrieves the transaction record, applies the dispute resolution policy, drafts the initial response to the customer, and flags cases that require manual investigation or regulatory reporting
- Client support summary worker — produces a periodic account support summary for customer success managers, showing open cases, resolution times, recurring issues, and sentiment signals extracted from case notes
| Worker Use Case | Research, Review, and Synthesis |
| What it does | A person needs to understand a topic quickly: a market, a competitor, a regulation, a body of academic literature, or an internal subject. Rather than manually reading through ten, twenty, or fifty source documents, an AI worker ingests the material, identifies the key themes and findings, extracts the important differences between sources, surfaces contradictions or knowledge gaps, and produces a structured summary or decision memo that the person can act on. |
| What triggers it | A research request, a competitive intelligence task, a pre-meeting briefing request, or a document uploaded for synthesis. |
| What it produces | A structured synthesis document: a literature review summary, a market landscape overview, a competitive brief, or a decision memo, with citations pointing back to the source material. |
| Where the human sits | The analyst, manager, or researcher reviews the synthesis, validates the source quality, applies judgment about what the evidence means for their specific decision, and decides whether the synthesis is complete enough to act on. |
| Where it shows up | Strategy teams preparing market analyses, research teams reviewing academic or clinical literature, sales teams building competitive intelligence, policy teams reviewing regulatory changes, and consultants preparing client briefings. |
| Why it matters | This is the most universal pattern in the library. Collect sources, compare them, condense them, recommend. Every knowledge worker does this regularly. The AI does not replace the judgment required at the end. It eliminates the reading time required at the beginning. |
Examples
- Literature review planner — ingests a set of academic or industry papers, extracts key findings, identifies themes and contradictions across sources, and produces a structured literature review outline for the researcher to develop
- Turn transit time into research time — captures voice notes, web pages, or document snippets gathered on the move and converts them into structured research notes with source citations
- Research to presentation assistant — takes a completed research synthesis and structures it into a presentation outline with suggested talking points, data highlights, and visual recommendations
- Cross-source market research synthesiser — aggregates analyst reports, news sources, competitor websites, and internal data into a single market landscape summary with key trends and implications identified
- Raw data statistics verification worker — checks statistical claims in draft documents against the original source data, flags unsupported assertions, and suggests corrected figures with citations
- Competitive intelligence briefing worker — monitors competitor activity across news, job postings, product announcements, and public filings, and produces a periodic briefing for strategy and sales teams
| Worker Use Case | Knowledge Reuse and Memory Reconstruction |
| What it does | Organisations consistently rebuild knowledge that already exists somewhere inside them. A sales team rebuilds competitor research Marketing completed six months ago. A new project lead reconstructs a methodology their predecessor documented and filed away. An analyst re-summarises a regulation that Legal reviewed two years ago. An AI worker searches across internal repositories, identifies what already exists, extracts the relevant content, and presents it in a form the current user can immediately apply, without them knowing where to look or even knowing the knowledge exists. |
| What triggers it | A new project kickoff, a repeated research request, a knowledge gap identified during a workflow, or a new team member needing to get up to speed on an existing domain. |
| What it produces | A reconstructed knowledge brief drawing from internal sources, with citations to the original documents so the user can go deeper if needed. |
| Where the human sits | The person receiving the brief validates whether the retrieved knowledge is still current, accurate, and applicable to their specific context. Knowledge retrieval does not replace judgment about relevance. |
| Where it shows up | Consulting firms and professional services organisations, sales teams with large product or competitive knowledge bases, research teams working on multi-year programmes, and any organisation that has been running long enough to have meaningful institutional memory buried in its systems. |
| Why it matters | Most organisations frame this as a content creation problem and buy more AI writing tools. The actual problem is memory reconstruction, and the infrastructure required including federated search, consistent metadata, and knowledge graphs is fundamentally different from the infrastructure required to generate new content. This card names that distinction explicitly. |
Examples
- Battle card library builder — retrieves existing competitive intelligence from across internal repositories, structures it into a consistent battle card format, and keeps it updated as new information is added to connected systems
- Customer persona builder — aggregates CRM data, support case themes, survey responses, and sales call notes into structured customer personas for use across marketing, product, and sales teams
- Program toolkit generator — recovers existing programme documentation, templates, frameworks, and lessons learned from previous projects and assembles them into a reusable toolkit for a new programme team
- Workflow improvement planner — analyses existing process documentation, support tickets, and team feedback to identify workflow inefficiencies and produces improvement recommendations with supporting evidence
- Institutional knowledge search assistant — provides a natural language search interface across internal repositories, returning relevant documents with context summaries so users find what exists without knowing exactly where to look
- Course and training content search worker — enables employees to search across thousands of internal learning resources using natural language queries and surfaces the most relevant materials matched to their role and current development need
| Worker Use Case | Meeting Intelligence and Action Capture |
| What it does | A meeting happens, live or recorded, and the AI worker handles everything that follows it. It transcribes the conversation, identifies speakers, extracts the decisions made and the actions assigned, notes the open questions that were not resolved, produces a structured summary, and pushes the actions into the relevant systems: the project tracker, the CRM, the task manager, or the team’s shared workspace. The result is that the meeting produces a complete, accurate record without anyone spending time after the call writing notes or chasing down who agreed to do what. |
| What triggers it | A meeting recording becoming available, a live meeting transcript completing, or a calendar event marked as concluded with a recording attached. |
| What it produces | A meeting transcript, a structured summary of key discussion points, a decisions log, an action item list with owners and suggested due dates, and open questions flagged for follow-up. Where connected systems are available, actions are logged directly into project or CRM tools. |
| Where the human sits | The meeting organiser or a designated participant reviews the summary and action list before it is distributed or logged, corrects any misattributions or misunderstandings from the transcript, and confirms action ownership with the relevant people. |
| Where it shows up | Executive and leadership teams running high volumes of decision meetings, sales teams logging client call outcomes into CRM systems, project teams tracking decisions across long programmes, and cross-functional teams where meeting accountability is a persistent problem. |
| Why it matters | Calendar, Project, and Status Coordination covers time-based coordination — preparing for what is coming. This card covers what happens after the moment passes. Together they form a complete coordination loop: brief before, capture after, track in between. Meeting intelligence is one of the most widely deployed AI patterns in practice precisely because it solves a problem every knowledge worker recognises immediately — the gap between what was agreed in a meeting and what actually gets recorded and acted upon. |
Examples
- Meeting summary worker — transcribes a recorded or live meeting, identifies speakers, and produces a structured summary of key discussion points and decisions reached
- Action item capture assistant — extracts assigned actions from a meeting transcript, identifies the owner and any stated deadline, and formats them into a task list ready for import into a project or task management tool
- CRM update from call worker — takes a sales call transcript and automatically logs the outcome, next steps, and account notes directly into the CRM, removing manual post-call data entry
- Compliance audit trail generator — creates an immutable, searchable record of what was said in regulated interactions such as financial advisory calls or sensitive negotiations, for compliance officer review
- Decision log generator — extracts formal decisions made during a meeting, timestamps them, and appends them to a running project decision record for future reference
- Follow-up communication drafter — takes the action items captured from a meeting and drafts reminder communications to the relevant participants, with a financial services firm already running this as a live agentic workflow