Enter your search text in the box above

Select a result to preview

AI Agents, Power Platform and SQL for Next Gen Automation

#fabcon2026 #session #learning #microsoft #fabcon2026-session

Description

Key Takeaways

My Notes

Action Items

Slides

Sound off.
The mic is all yours.
Influence the product roadmap.
Join the Fabric User Panel
Join the SQL User Panel
Share your feedback directly with our
Fabric product group and researchers.
Influence our SQL roadmap and ensure
it meets your real-life needs
https://aka.ms/JoinFabricUserPanel
https://aka.ms/JoinSQLUserPanel
WELCOME
AI Agents, Power
Platform and SQL
for Next Gen Automation
WHY YOUR AUTOMATION STRATEGY NEEDS TO EVOLVE NOW
INTRODUCTION
Hello,
Chike Eduputa
Head of Microsoft
Capgemini Invent UK
in/ceduputa
AGENDA
In this session, we will cover:
The Treadmill Problem
The quiet hero
What is an Agent
Trusted systems
Where do Agents fit
CHAPTER 1
The Treadmill Problem
CHANGE OF CUSTOMER DETAILS PROCESS
Gather
required
identification
documents.
Visit a branch.
Some companies
accept forms by Post
or Online
Complete a
Paper form.
Some companies
use Digital Forms
Verify identity
with relevant
identification
documents.
Receive
confirmation
of updated
details.
Sounds simple and straightforward to automate, right?
CHAPTER 1
The Treadmill Problem
CHANGE OF CUSTOMER DETAILS PROCESS
Customer identifies
the need to update
personal details.
Visit a branch, queues and
speaks to a staff (typical). Some
banks accept by Post or Online
Complete a Paper form.
Some banks have a
Digital Form
Verify identity with
relevant identification
documents.
Receive
confirmation of
updated details.
Paper form and ID are batch
scanned and manually
entered in internal systems
Branch Teller
System used by staff
to capture updates
Data entry team
reviews for legibility
and completeness
Document Scanning
System to digitise
paper forms
Identity
verification checks
are performed
Customer Information
File to master
customer profile
Business rules applied like
fraud screening, address
validation, duplicate records
Core Banking
System for Account
Servicing
Fraud Detection
System for risk
screening
Customer
information file
updated
Changes propagated
to core banking
systems downstream
Address Verification
Service to validate
addresses
Data Integration
system to sync updates
across systems
Audit logs are
created for
compliance
Notification sent
to customer
confirming update
Notification service
to send SMS, email
and/or post
Audit and
Compliance System
for regulatory tracking
CHAPTER 1
The Treadmill Problem
BANK CHANGE OF CUSTOMER DETAILS
Customer identifies
the need to update
personal details
Complete a Paper form.
Visit a branch, queues and
Some banks have a
speaks to a staff (typical). Some
Digital Form
banks accept by Post or Online
Requires physical
Queues and
branch visit
waiting times
Paper form is batch
scanned and manually
entered in internal systems
Branch Teller
System used by staff
to capture updates
Data entry team
reviews for legibility
and completeness
Paper handling
and storage
Document Scanning
System to digitise
paper forms
Identity
verification checks
are performed
Customer Information
File to master
customer profile
Business rules applied like
fraud screening, address
validation, duplicate records
Manual data
entry errors
Core Banking
System for Account
Servicing
Fraud Detection
System for risk
screening
Verify identity with
relevant identification
documents.
Receive
confirmation of
updated details.
Manual form
errors
Slow turnaround
times
Customer
information file
updated
Changes propagated
to core banking
systems downstream
Address Verification
Service to validate
addresses
Data Integration
system to sync updates
across systems
Slow processing
times
Audit logs are
created for
compliance
Notification sent
to customer
confirming update
Notification service
to send SMS, email
and/or post
Audit and
Compliance System
for regulatory tracking
Fragmented
system updates
CHAPTER 1
The Treadmill Problem
Now picture a 100+ years old Global Bank,
100,000’s
1,000’s
10,000’s
10,000’s
Transactions per year
Form types
Queues
Operations Staff
Paper forms
Manual entry
AI Builder + Power Automate Desktop struggled
FRAGILITY
CHANGE FAILURE
The pixel perfect fallacy

  1. Trigger
  2. Extraction Agent
    Email received with scanned
    handwritten PDF Form
    Azure Doc Intelligence
    Process Orchestration Agent
    Copilot Studio
  3. Process #1 Change of Name Expert
  4. Process #2 Direct Debit Expert
  5. Exceptions Agent
  6. Customer Insights
    Copilot Studio
    Copilot Studio
    Copilot Studio
    Fabric Data Agent
    Query customer history
    3.1
    3.2
    3.3
    4.1
    4.2
    4.3
    5.1
    Details
    Check
    Fraud
    Check
    Update
    CIF
    Details
    Check
    Send to DD
    Work Queue
    Update
    CIF
    Human Check
    Human in the Loop
    via SQL / Fabric
    Customer Information File / Microsoft Fabric
    DEMO
    Solution Demo
    CHANGE OF CUSTOMER AGENTIC SOLUTION
    CHAPTER 4
    Microsoft Fabric AI
    DATA AGENTS
    COPILOT IN FABRIC NOTEBOOKS
    AI FUNCTIONS
    AI agents that converse with your
    OneLake data, enforce governance, and
    surface actionable insights
    In-cell AI generates code, SQL
    completions, and data transformations
    across all Fabric workloads
    LLM-powered summarization,
    classification, and text generation on
    OneLake data in a single line of code
    Agent queries customer history across lakehouse
    tables to validate address change requests
    Copilot writes the PySpark to detect anomalous
    address changes across customer records
    Classify change requests as routine, suspicious, or
    high-risk using AI functions over lakehouse data
    CHAPTER 4
    Power BI Copilot
    NATURAL LANGUAGE INSIGHTS
    AI REPORT CREATION
    EMBEDDED EVERYWHERE
    Standalone Copilot finds and analyses
    any report or model. Ask questions, get
    instant visuals and summaries
    Copilot creates entire report pages,
    selects best visuals, and auto-generates
    DAX measures with descriptions
    Copilot in SharePoint, Teams, mobile
    apps, and org apps. Insights at every
    decision point
    "Show address changes by region this quarter"
    instantly generates a filtered dashboard
    Build a customer details audit report in seconds,
    with anomaly detection visuals auto-selected
    Service agent in Teams asks Copilot about a
    customer's change history mid-call via embedded
    report
    CHAPTER 4
    SQL Server 2025 AI
    COPILOT IN SSMS
    VECTOR SEARCH
    FABRIC MIRRORING
    Natural language to T-SQL in SSMS 21.
    Write, explain, fix, and optimise queries
    via chat
    Native VECTOR data type and
    VECTOR_SEARCH() for semantic search
    directly in T-SQL. No external DB needed
    Zero-ETL mirroring streams SQL Server
    data to OneLake in near real time for AI
    analytics without impacting OLTP
    "Show all customers who changed address more
    than twice this year" generates the T-SQL instantly
    Semantic search finds similar past address fraud
    cases by meaning, not just keyword matching
    Customer address changes in SQL Server are
    mirrored to Fabric for real-time anomaly analysis
    CHAPTER 4
    SQL is the quiet hero
    DETERMINISTIC ANCHOR
    VALIDATION LAYER
    CONTROL PLANE
    AI generates probabilistic outputs. SQL
    delivers exact, repeatable results every
    time. When precision matters, SQL is the
    final arbiter
    Audit the SQL, not the prompt. Stored
    procedures and views create an
    explainable, version-controlled logic
    layer
    Schema engineering replaces prompt
    engineering. Well-designed tables,
    constraints, and relationships define
    what agents can and cannot do
    An agent calculates a customer's exact account
    balance via SQL rather than estimating it from
    context
    A compliance team reviews the stored procedure
    an agent used, not the unpredictable prompt that
    triggered it
    Foreign keys prevent an agent from creating an
    order for a non-existent customer, no prompt guard
    needed
    CHAPTER 4
    SQL is the quiet hero
    STRUCTURED MEMORY
    TRANSACTIONAL SAFETY
    GOVERNANCE AT SCALE
    Agents need persistent, queryable
    memory. SQL tables give agents
    structured recall across sessions
    ACID guarantees mean agent actions
    either fully commit or fully roll back. No
    partial writes, no data corruption
    Row-level security, audit logs, and rolebased access built into SQL give agents
    guardrails enterprises already trust
    An agent recalls a customer's full order history
    from SQL before recommending next steps
    An agent processing a refund either updates the
    order, adjusts inventory, and logs the action, or
    none of it happens
    Agents can only access rows they are authorised
    for, with every query logged for compliance
    CHAPTER 5
    Trust in AI Agents
    The "Agentic Era" is here, marking a shift from
    chatbots that only converse to agents that
    take actions, invoking APIs, accessing
    databases, and overseeing their own longterm memory.
    However, increased autonomy introduces new
    risks. How can we ensure that agents stay
    secure, compliant, and behave as expected?
    Source: OWASP Top 10 for Agentic Applications 2026
    CHAPTER 5
    Trust in AI Agents
    Source: NIST AI RMF 100-1
    CHAPTER 5
    Trust in AI Agents
    Source: ISO/IEC 42001 Certification
    CHAPTER 5
    Trust in AI Agents
    Source: EU AI ACT Risk Levels
    CHAPTER 5
    Trust in AI Agents
    GOVERN
    MAP
    • Establish Security by Design
    culture across agent lifecycle
    • Identify full attack surface of
    each agent
    • Conduct Red Teaming for
    agentic-specific threats
    • Deploy guardrails and
    continuous monitoring
    • Define accountability chains:
    who is liable for unauthorized
    API calls?
    • Catalog: LLM, system prompt,
    tools (APIs), RAG sources
    • Assess Groundedness scores
    and hallucination rates
    • Prioritize: Excessive Agency
    (LLM08) Goal Hijack (ASI01)
    • Map agent-to-agent trust
    boundaries
    • Proactive stress-testing before
    deployment
    • Integrate Defender for Cloud +
    Defender for AI
    • Assess identity chain: Entra
    Agent ID, RBAC, service
    principals
    • Evaluate against OWASP
    ASI01–ASI10
    • Implement kill switches and
    circuit breakers
    • Evidence-based governance:
    policies, attestations, tollgates
    • Align with NIST AI RMF 100-1
    MEASURE
    MANAGE
    Source: Abhi Singh, Microsoft Defender for Cloud Blog 2026; Microsoft Architecting Trust Framework
    CLOSE
    Next Steps
    When you go back to work on Monday:
    Continue the AI
    Agents conversation
    Learn by building low
    risk, high ROI agents
    Promote trust and
    security by design
    CLOSE
    Thank you.
    How was
    the session?
    Complete Session Surveys in
    for your chance to WIN
    PRIZES!