Gokan's Studio

Orchestrate your knowledge (at scale?)

UPDATE: Public Preview Announcement (April 21, 2026)

I postponed writing the article until the private preview.  Today, at the Microsoft 365 Community Conference, Microsoft announced that AI Skills are now entering public preview with significant new capabilities, hence I think there is no better place to publish my first impressions.

Image Credit: Vlad Catrinescu, and if you want to learn more about Skills, please check his video too: https://www.youtube.com/watch?v=z_86-7kkpzw&feature=youtu.be

Skills got announced, and three major updates reshaped what was incomplete:

1. “What to Know” — Organizational Context:
   SharePoint now supports persistent context files at the site level. You can configure SharePoint to “remember” team preferences, brand colors, writing styles, and procedural rules. This context is stored in a Markdown file and automatically applied to all AI interactions on the site.
2. “How to Act” — Enhanced Skills Framework:
   Skills are now the core feature for multi-step, repeatable workflows. I am to call it “orchestrating the knowledge” Microsoft could come with another naming or people could disagree – but that’s how I see it. Anyone on a site can create, refine, and reuse skills for tasks such as generating quarterly reports, drafting proposals from templates, or applying content taxonomy.
3. “What to Produce” — Native Content Generation:
   SharePoint AI can now directly generate Office files—including Word documents, PowerPoint presentations, and Excel spreadsheets—as well as create structured artifacts such as dashboards and reports. These outputs can also be saved as reusable skills.

All these capabilities are currently rolling out to tenants enrolled in the public preview named Skills. To enable Skills in SharePoint, you only need to enable “Skill Assets” in the Site Collection features if you’re part of the preview program:

https://learn.microsoft.com/en-us/sharepoint/ai-in-sharepoint-get-started

Agent Assets Library: Centralized Governance

Skills and context files are now stored in a new, dedicated Agent Assets library on each site. This library is version-controlled, auditable, and governed by existing SharePoint permissions and compliance policies. Unlike ad-hoc skill creation, this establishes a clear home for organizational knowledge.

The architectural clarity is notable. The system now cleanly separates:

• Context (sharepoint.md): What the AI needs to know about your organization
• Skills (skill.md files): How the AI should act and what processes it should follow
• Output (generated files): What the AI produces and makes available for reuse

This pattern mirrors industry best practices emerging in Claude Skills and other AI agent frameworks. It’s architecturally sound and positions SharePoint to evolve independently of individual feature announcements.

Let’s create our first Skill

Microsoft Community Conference April 2026

Here is a simple flow for creating a Skill in SharePoint. Open the AI in SharePoint Agent (AKA the knowledge agent) – and write “create me a skill:” and insert your set of instructions that comes through Copilot, Claude Code, or any other LLM.

AI in SharePoint will propose a Skill name, its Instructions (because, at the end, an MD file is a collection of instructions), and create the Skill once you confirm.

By using //skills you’d be able to see all skills you’d have in your site, select the newly created skill in run it.

Once the skill loaded, you’d be able to see the result in the Agent Pane.

The good part: AI Skills remain a genuine leap forward

The Skill creation experience is genius, simple.

Creating an AI Skill in SharePoint requires almost no technical knowledge. Write a Markdown file describing what the skill should do. Paste it into the Agent. Done. The simplicity is a feature, not a limitation. With public preview, this experience is even better: you can create and edit skills directly in the browser, ask Copilot to help you write skill Markdown, and iterate without ever leaving SharePoint.

The Power is real when it works

The ability to move documents based on conditional logic, applying consistent metadata using organizational taxonomy, draft proposals from templates, and generate quarterly reports—all without writing code—remains genuinely transformative for power users, departmental administrators, and small organizations.

In the Microsoft 365 Community Conference sessions this week, multiple organizations shared use cases in which skills compressed what would otherwise require Power Automate flows or PowerShell scripts into minutes of natural-language interaction.

The Future part: AI Skills remain a genuine leap forward

Improvement request #1: Hard Boundaries on External Systems

Skills cannot connect to external systems or execute custom code. A skill can only perform actions that the user already has permission to do. This means:

• Cannot directly integrate with Fabric, Foundry, ServiceNow, Salesforce, Jira, SQL databases, or other Power Platform apps like Power Automate or Power Apps
• Cannot connect to Dataflows or Dataverse
• Cannot create tasks in Microsoft Planner directly (requires Power Automate connector)

For enterprises integrating multiple systems, skills are a complementary layer rather than a replacement for Power Automate. The limitation is clear and honest.

Improvement request #2: Tenant-Level Skill Hub and Catalog

Organizations need a way to publish approved skills at the tenant or hub level and have individual sites subscribe to them, rather than manually copying skill files. In other words: A tenant-level skill hub that allows IT to publish approved, curated skills across all sites with categorization, ratings, and usage analytics. Cross-site skill sharing remains manual.

This should include:

• Skill marketplace model with categorization and description standards.
• Versioning and rollback capabilities on the tenant level.
• Usage analytics showing which skills are adopted and how often they are invoked
• Approval workflows for IT-curated skills

Improvement request #3: Purview Integration for Audit Trails

Skills that take actions on content must log to Microsoft Purview. This is non-negotiable for regulated industries and required for GDPR, HIPAA, SOX compliance.

Audit entries should include:

• Actor (user who invoked the skill)
• Timestamp
• Skill name and version
• Items affected and actions taken

Improvement request #4: Power Platform Integration

While skills today cannot be directly connected as triggers or actions in Power Automate, this limitation will frustrate enterprise architects. A secure, pre-approved set of connectors would extend the use case without introducing unlimited risk. It would be fantastic if we could call a Power Automate flow via a Skill or vice versa.

Improvement request #5: Skill Lifecycle Governance

Today, Skills are introduced and attached at the Site level, but as mentioned before, a Skill should be able to be attached to a person, process, hub site, SharePoint home, or even a group. Introduce formal governance for skills:

• Draft (creator is testing)
• Submitted for Review (waiting for approval)
• Approved (can run on production content)
• Deprecated (discouraged, marked for removal)

Improvement request #6: Attached Skills and Cross-Organizational Knowledge

Formalize the attached skill pattern to ensure expertise persists across organizational boundaries. This might include:

• Metadata fields to indicate the expert who created and maintains a skill
• Access controls that allow experts to maintain skills even after moving roles
• A skill manifest or registry that tracks ownership and allows discoverability of expert-maintained skills

Improvement request #7: Attached Skills – Knowledge Walks Out the Door

Let’s say Gokan is a compliance expert. He’s spending weeks building expertise in contract review, regulatory frameworks, and company policy. He does understand risk thresholds, knows which clauses matter, and can spot problems in seconds. Then Gokan gets promoted to executive and moves to a different department, team, and reporting structure.

What happens to my expertise, to my skill, to my knowledge? In a traditional organization, it walks out with Gokan. Gokan should be able to mark a skill, so it follows him wherever he goes (promotions, team changes, job changes, …).

Improvement request #8: Preview

Before executing a Skill, a preview (pane) would be beneficial. While we know that destructive actions are human-validated and controlled, there is no undo action in SharePoint; hence, not on skills either. Executing a skill with incomplete logic on corporate data could have a dramatic impact on data quality.

Where are SharePoint AI Skills today?

April 2026

Here is the updated assessment of SharePoint AI Skills against enterprise readiness criteria:

Capability	Status (Public Preview)	Enterprise Ready?
Skill Creation	Simple Markdown in browser	✓ Yes
Skill Discovery	/skills command + Agent Assets library	⚠ Partial
Permission Control	Inherit from site; break for granular control	✓ Yes
Audit Trail	No integration yet	✗ No
Content Type Awareness	Manual validation in skill design	⚠ Partial
Org Context (sharepoint.md)	Store at site level; applies to all users – not tenant level	⚠ Partial

My thoughts

April 2026

The Vision Is Right. The Execution Is Improving.

When I first looked at AI Skills in private preview, I saw a genuinely innovative concept that still had room to grow for enterprise use. Public preview moves things forward in a meaningful way. The addition of organizational context, the Agent Assets library, native browser editing of Markdown, and the clear separation between know, act, and produce all point to thoughtful product design.

This shows that Microsoft is building this with real care and a strong architectural direction. Public preview is not the same as full enterprise maturity, though, and that is perfectly okay. There are still a few capabilities that would make the experience even stronger for larger organizations, such as tenant-level skill governance, approval workflows, Purview integration for audit trails, and native content type or metadata validation.

For individual users and departmental power users, this is a great time to start experimenting, learning, and building. The experience is solid, and the potential is exciting. For enterprise architects and CIOs, the right mindset is one of informed optimism: skills are a promising foundation, while the governance layer continues to evolve. In the meantime, it makes sense to plan for manual reviews, permission controls for sensitive content, additional audit tracking where needed, and clear business cases for each skill.

The future of SharePoint is clearly becoming more agentic, and AI Skills are an important part of that direction. The foundation is strong, the momentum is encouraging, and the next layer is still being built.

Hope this helps!

Renewed Revolution!