Andrew Connell hosts his monthly office hours covering recent updates in the Microsoft 365 development ecosystem. The session focuses on SharePoint Framework (SPFx) version updates, the Microsoft 365 Agents Toolkit releases and quality issues, new Copilot capabilities for developers, and the announcement of skills in SharePoint. Andrew also highlights his upcoming three-day workshop on building declarative agents and discusses his philosophy on using AI as a development augmentation tool.
- 01:45 - Welcome & Session Overview
- 04:32 - SharePoint Framework Updates
- 09:33 - SPFx v1.23 Features
- 13:52 - Microsoft Teams & Agents Toolkit
- 19:06 - Copilot New Capabilities
- 25:22 - Declarative Agents Workshop
- 31:09 - M365 Community Conference News
- 39:08 - Skills in SharePoint
- 44:02 - AI-Augmented Development
- 56:36 - Closing & Next Steps
Key Points
- SharePoint Framework v1.23 Release: SPFx v1.23 is in release candidate stage with planned GA by mid-April, featuring grouping support for list view command sets and a new CLI tool (preview) to replace the Yeoman generator
- M365 Agents Toolkit (ATK) for VS Code Quality Issues: Recent versions (6.6, 6.8) contain multiple bugs including JSON formatting errors, schema generation issues, and filtering problems that require careful attention
- New Copilot Capabilities: Work IQ CLI plugin and agent plugins for declarative agents enable integration with AI coding assistants; UI Widget Developer allows MCP servers to return React components rendered inline in Copilot
- Skills in SharePoint (Preview): Microsoft announced preview support for skills in SharePoint, allowing context-driven agent behaviors similar to capabilities in Claude, ChatGPT, and GitHub Copilot
- AI-Augmented Development: Andrew shares his shift toward letting AI generate the majority of code while he focuses on planning, reviewing, and editing—a development manager role rather than hands-on coding
- Upcoming Workshops: Three-day workshop on building declarative agents running April 28-30 with live lectures, instructor-led demos, and hands-on assignments covering JSON/YAML, TypeSpec, SharePoint lists, and UI components
News & Links Referenced
Detailed review of Microsoft 365 Agents Toolkit v6.6.0: GCC-M support, embedded knowledge for declarative agents, MCP GA, and 61 templates removed.
https://www.voitanos.io/blog/microsoft-365-agents-toolkit-v6-6-0-release-review/
SharePoint Framework roadmap update as of late March 2026
https://devblogs.microsoft.com/microsoft365dev/sharepoint-framework-spfx-roadmap-update-march-2026
SPFx v1.23 release notes from the SharePoint developer documentation.
https://learn.microsoft.com/sharepoint/dev/spfx/release-1.23
Starting from the SharePoint Framework v1.23, ListView Command Sets also support grouping. You can use group element and cross item associations in the ListView Command Set manifest to build a hierarchy of commands that are then grouped in the list and library toolbar and context menu.
https://learn.microsoft.com/sharepoint/dev/spfx/extensions/guidance/list-view-command-set-grouping
MCP Server and CLI for accessing Work IQ
https://github.com/microsoft/work-iq
MCP Apps, embedded knowledge, and the M365 Agents Toolkit plugin just changed declarative agent development. Here's what developers need to know.
https://www.voitanos.io/blog/microsoft-365-copilot-declarative-agents-whats-new-202604-april-2026/
I've been using AI agents in my daily development workflow for a few months, and what I'm doing looks nothing like what most people mean by vibe coding.
https://www.voitanos.io/blog/vibe-coding-vs-agentic-engineering/
Patterns for getting the best results out of coding agents like Claude Code and OpenAI Codex.
https://simonwillison.net/guides/agentic-engineering-patterns/
Chris McNulty interviews Andrew Connell about AI-assisted software development, vibe coding, and the future of AI in enterprise tech.
https://www.youtube.com/watch?v=47QJx6cGmD0
SharePoint Framework Updates
The SharePoint Framework ecosystem is relatively quiet, with most updates being maintenance-focused. Andrew references the monthly status article (posted ~end of March) that covers current developments.
Current Version Status
- Currently on SPF v1.22.2
- Version 1.23 is in release candidate stage with planned general availability target of April 15 (though that date has passed and a regression was found)
- A breaking change regression was discovered on April 26 involving CSS hash changes in global selectors within CSS/SASS modules
- The regression involves the 1.23.1 version hashing CSS classes when using global selectors—a behavior that should not occur and will likely be fixed before GA release
SPFx v1.23 Features
Grouping Support for List View Command Sets
Command sets now support grouping, allowing subgroups of commands beneath a parent group. This is implemented through a simple configuration approach where you define a group and assign commands to that group via a group property.
New CLI Tool (Preview)
- Intent: Replace the Yeoman generator for SharePoint Framework project creation
- Status: Currently in preview; full GA expected around June/July 2026
- Strategy: Will be available side-by-side with Yeoman generator; existing projects don’t require immediate migration
- Features: Custom templates supported via GitHub repo; public template repositories; ability to define custom templates
- Command format:
spf createwith prompts similar to existing Yeoman workflow
Microsoft Teams & Agents Toolkit
Recent releases of the Agents Toolkit (6.6 in mid-March, 6.8 in late April) introduced significant updates but also quality concerns.
Agents Toolkit Version Updates
- 6.6 Release: Focused on cleanup and maintenance items
- 6.8 Release: Added template modifications and numerous changes
Quality Issues Identified
Andrew found multiple bugs in the 6.8 release during his workshop at the M3 65 Community Conference (the toolkit updated 42-43 minutes before his demo). Issues discovered include:
- JSON task file generation with missing commas, creating invalid configuration files
- TypeSpec-based agent generation dropping properties when using internal schema generation tools
- Documentation errors
- Filtering capability issues with OneDrive/SharePoint in declarative agents
- Additional issues discovered post-6.8 release still pending PR fixes
Several bug reports have been submitted and are in various stages of resolution (some already fixed, others still open). The quality inconsistency is frustrating given the criticality of this toolkit for building declarative agents.
Copilot New Capabilities
New Copilot capabilities announced by Microsoft demonstrate advanced integration patterns for developers building with declarative agents.
Work IQ Plugin
- Conceptually similar to Microsoft Graph but adds contextual awareness and memory
- Includes CLI plugin for direct access
- Agent plugin available for AI coding assistants (GitHub Copilot, Claude Code, Google Gemini)
- Enables declarative agents to leverage Microsoft Graph capabilities with additional context about users and experiences
Agent Plugins for AI Coding Assistants
Collections of skills/resources for working with declarative agent definitions (JSON, YAML, text files). These plugins help apply best practices and guide agent behavior configuration without requiring TypeScript or TypeSpec.
UI Widget Developer (MCP Extension)
A powerful new capability allowing MCP servers to return UI elements rendered inline in Copilot:
- Enables React components/forms to be rendered directly in the chat interface
- HTML and dependencies sent from server to Copilot and rendered in an iframe within Copilot’s security boundary
- Allows data passing between UI component and chat experience
- Provides alternative to pure chat-based interaction when UI/forms are more appropriate
- Uses MCP App specification (official extension to MCP spec) rather than legacy OpenAI App SDK or MCP UI specs
- MCP App spec works cross-platform (Claude, Copilot, etc.) whereas OpenAI spec is ChatGPT-only
Declarative Agents Workshop
Andrew is running his sixth cohort of a three-day workshop on building declarative agents (April 28-30, 2026). The workshop combines live instruction with self-paced demonstrations.
Workshop Format
- Two 90-minute live lecture sessions per day (explanation, teaching, live demos)
- 8-10 instructor-led demo videos (self-paced step-by-step walkthroughs)
- All content recorded; live sessions posted after delivery
- Q&A through chat with responses usually posted same day
- Code samples and downloadable content included
- Recording access extends through June for catch-up
New Additions to This Cohort
Three new optional homework assignments/demos:
- SharePoint List Integration: Read and write SharePoint lists using declarative agents without custom code—uses Microsoft Graph REST API directly through agent configuration (three endpoints: two GETs, one POST)
- UI Widget Integration: Extend a PTO request agent to include the UI Widget Developer capability for form-based interaction instead of pure chat
- TypeSpec Implementation: Build the same PTO request agent using TypeSpec instead of JSON/YAML for a more robust developer experience
Pricing & Accessibility
- $349 per person standard pricing
- Special pricing for unfavorable exchange rate countries (10-60% discount for ~60 supported countries)
- Automatic geo-detection at checkout; manual support available for verification
Registration Details
- Curriculum broken down by session on website
- Each calendar session must be imported individually (6 total)
- Early bird discounts already claimed
M3 65 Conference News
The M3 65 Community Conference took place April 19-25, 2026 in Orlando with two pre-conference workshop days, three days of breakout sessions, and one post-conference workshop day. Andrew presented a full-day workshop on declarative agents and delivered breakout sessions on Copilot and SharePoint Framework.
Skills in SharePoint (Preview Announcement)
The primary developer-focused announcement from the conference is the preview release of skills in SharePoint. This feature enables context-driven agent behaviors similar to capabilities in Claude, ChatGPT, and GitHub Copilot.
Skills Concept
Skills teach agents how to behave in specific scenarios without explicit instruction every time:
- Agents don’t explicitly call skills; rather, skills are available based on name and description
- When scenarios match skill descriptions, agents automatically leverage the skill context
- Think of it as “coaching the agent” on preferred behaviors and outcomes
Andrew’s Skills Examples
Voice Guidelines: Analysis of his top 10 articles to capture speaking patterns, writing style, key phrases, and avoiding conventions (e.g., em dashes). Used when refining AI-generated content to maintain his voice.
Brand Kit: Company brand guidelines, colors, logo usage, company reference standards, ensuring consistency in AI-generated materials
Glossary: Term definitions and formatting (e.g., “Microsoft 365” on first mention then “M3 65”; “SharePoint Framework” becomes “SPF”; “Entra ID” proper casing)
Office Hours Processing: Skill to extract summaries, chapter points, internal links, and notes from office hours transcripts in his voice
SharePoint Skills Use Case
When users request list creation, skills define the preferred configuration approach without repeated instruction. For example, a skill might specify that PTO request lists should reference person objects by ID rather than email, automating best practice application across multiple interactions.
AI-Augmented Development Philosophy
Andrew emphasizes using AI as a productivity multiplier rather than a replacement for developer expertise. His approach evolved significantly since late 2025 with the emergence of more capable tools.
Core Philosophy
- AI should augment capabilities, not replace expertise
- Analogy: Using tools (drills, saws, calculators) requires understanding to use them correctly
- Misuse creates frustration and failure
- Not suited for: Asking AI to solve problems you don’t understand without verification
Two Approaches: Vibe Coding vs. Agentic Engineering
- Vibe Coding: Running code generation without review; useful for experimentation and rapid iteration but risky without expertise
- Agentic Engineering: Structured, planned approach using AI to generate code against defined specifications with review and refinement
Both have value depending on context; Andrew has shifted appreciation toward vibe coding for rapid experimentation while maintaining oversight.
Andrew’s Workflow Evolution
Over the last two months, his hands-on coding percentage has dropped dramatically:
- Only ~5% hand-written code recently
- ~95% AI-generated with his oversight
- New role: Development manager vs. hands-on engineer
- Process: Define plan/tasks/outline → AI generates → Review → Edit → Deploy
Lab Demo Construction Workflow
For his workshop lab demos, Andrew uses an AI-augmented multi-stage process:
- Create master project with all functionality working
- Break down into bullet-point steps and demo markers
- Create fresh project and commit each step, dictating commit messages
- Create third project and use AI agents/skills/plugins to:
- Research entire project and git history
- Map out structure as “play” (big demo) → “acts” (sections) → “scenes” (steps)
- Generate complete step-by-step demo instructions
- Creates click-through demo flows without typos or figuring out implementation details in real-time
Industry Perspective
- CEO of Anthropic and Boris Kenney (Claude Code developer) have stated that the “software engineer” job title will likely disappear within 1-2 years
- Andrew increasingly agrees with this assessment based on his tool usage patterns
- Message to engineers: If you’re not using these tools, someone using them will have significant productivity advantages
- Frames AI as another layer of abstraction evolution (assembly → C++ → Java/managed languages → scripting languages → AI-augmented development)
Closing & Next Steps
Upcoming Engagements
- Three-day declarative agents workshop this week (April 28-30)
- European Collaboration Summit in Cologne, Germany next week (May 3-5 approximately) delivering full-day workshop and breakout sessions on Copilot and SharePoint Framework
- Workshop with Mark Rackley at European conference: “Iron Chef” style comparison building the same agents with different tools (Mark using Power Apps Studio, Andrew using Agents Toolkit)
Free 9-Day Email Course
