S1.1 Why a company assistant rather than a free tool

• How Shadow AI works, with concrete examples
• What the user gets in return
• Data admissibility rules: “allowed / conditional / not allowed”
• Corporate sign-in (SSO)

S1.2 Choosing an assistant and the anatomy of a conversation

• The assistant picker, layouts, your own ordering
• Why there are several assistants rather than one
• Choosing a model mid-conversation and the cost–quality trade-off
• Four working modes: extended reasoning, internet, short answers, Deep Search
• Stopping and resuming an answer
• Resilience to an interrupted session
• Editing a question and answer variants
• A visible trace of the assistant's work
• Keeping conversation history in order
• Copying, saving formulas and amounts, light/dark mode, interface language, message length limit

S1.3 Practical prompting

• The skeleton of a good instruction
• Iteration instead of a single shot
• Anti-patterns
• What you do not need to type
• Answers picked from a list
• The prompt as a team asset; nominating candidates for Skills
• Exercise: your own end-to-end task, timed before and after

S1.4 Verifying answers and the limits of trust

• Where hallucinations come from
• Reliable and unreliable tasks
• Reading source references and the excerpt preview
• Three markers: source used, source found but unused, the model's general knowledge
• Platform safeguards: relevance checking, no quoting from memory
• The rule: you sign it with your name — you check it
• Reporting a bad answer to the tenant owner

S1.5 Output, integrations, token budget

• Output files: documents, spreadsheets, presentations, images
• The connected accounts panel — linking your own account and its permission scope
• Approving irreversible operations
• Invoking Skills
• Budget: warning, block, how the model and mode affect consumption
• Awareness of conversation oversight

S1.6 Attachments in a conversation

• Only for assistants permitted to work with files
• An ad-hoc file in a conversation versus a document in the knowledge base

• How a knowledge base works and what that means for how you phrase a question
• What the platform does automatically, and what the user does not have to do
• Narrow versus broad questions; questions about completeness
• Filtering by descriptive attributes
• When to switch on Deep Search
• Working with a long document
• Numerical data and tables
• Spotting a gap in the base and reporting it to the owner

S3.1 Designing an assistant and dividing work into tenants

• From task to assistant: an assistant, a Skill or a prompt
• The assistant card
• Tenant boundaries: confidentiality, knowledge ownership, billing
• Assistant groups and permissions
• The assistant lifecycle; the limit on the number of assistants
• Visual identity and the welcome message

S3.2 Creating assistants

• The library of ready-made assistants
• When to use the library, when to use the Wizard
• The assistant Wizard — a full walkthrough
• The draft prompt for approval
• The Wizard's safety rules
• What the Wizard does not know — the owner's remaining work

S3.3 System prompt — workshop

• The structure of a prompt
• The hierarchy of knowledge sources
• The balance: prompt / knowledge base / Skill / model
• Boundaries of responsibility
• Rules that limit hallucinations
• Tone and answer format
• Prompt versioning

S3.4 The knowledge base — the owner's work

• Qualifying a document for the base
• Adding documents, statuses, reprocessing
• OCR for scans
• Descriptive attributes and bulk completion
• Designing an attribute set — an exercise on the client's own corpus
• Document structure and navigating by headings
• Spreadsheets: preparation and analysis
• Configuring Deep Search per assistant
• Maintaining the base as an ongoing process

S3.5 Skills — when and what for

• What a Skill is and why it exists
• The anatomy of a Skill
• Deliberate and automatic invocation
• Executable Skills and the sandbox
• Bulk import of Skills
• Assigning Skills to assistants
• Exercise: building a Skill

S3.6 Models, tools and permissions within an assistant

• Assigning models and per-model features
• ZDR and the EU region requirement per model
• The minimum tool set
• Three operation modes: allowed / requiring approval / blocked
• Access to external information and address allow-lists
• Formats of generated material and their limits
• Permissions down to a single configuration field

S3.7 Quality, cost and maintenance

• Diagnostics: “what to do when the result is poor”
• Oversight of conversation quality
• Assistant analytics
• Cost optimisation based on data
• Maintaining an assistant

S3.8 Closing workshop

• Building a departmental assistant from scratch to a test set, with a demo for participants

• The model catalogue
• ZDR and the EU region requirement
• Tenants and their lifecycle
• Users and accounts
• Roles built from elementary permissions — exercise: three intermediate roles
• User groups
• Corporate identity and SSO
• Active Directory / Microsoft Entra ID; onboarding and offboarding
• Organisation and assistant-group analytics
• Billing by operation type
• Limits and balance; ordering tokens
• Background work and scheduled tasks
• White-label

• The performance-quality settings panel
• A method for tuning parameters
• MCP — what it is and what it is for
• Connecting an MCP server
• Microsoft 365 — the configuration wizard and permission mapping
• Zoho — CRM, mail, calendar, processing regions
• Atlassian — Jira, Confluence, Compass
• ServiceDesk Plus — permission control and audit logging
• The sandbox for executable Skills
• Integration diagnostics
• Integration hygiene

• Assistant versus agent
• Integration with external data; MCP as the route to industry systems
• How to recognise a process suited to an agent
• Human-in-the-loop
• Safeguards and operating boundaries
• What questions should I be asking — workshop
• Disclosure obligations and accountability
• From idea to deployment with PLC