Opal in Q2 2026: From Assistant to Operating Layer

Q2 2026 was the quarter Opal stopped being a chat window and started behaving like infrastructure. Across April, May and June, Optimizely shipped a browsable agent catalog, an automated quality layer for agent runs, EU data residency, and a skills system that lets teams encode their own brand voice and compliance language once and reuse it everywhere. Here is what landed, and what it means for how your team works.

Discovery: the Agent Library and Personalize Opal

The single biggest barrier to Opal adoption was never capability, it was knowing what existed. June's Agent Library fixes that with a browsable catalog on the Opal Chat homepage where more than 45 pre-built agents can be found and launched in one click. Alongside it, Personalize Opal is a pinned onboarding agent that walks a new user through connecting their tools and building their first skills in a guided, role-based session, so value arrives in the first few minutes rather than the third week.

Quality: scoring, guardrails and recovery

Scaling agents across an organisation only works if you are not manually reviewing every execution. The Quality tab for specialized agents, released 1 June, introduces three capabilities that make that possible:

• Output Evaluation — an AI model scores each run against up to 10 criteria you define, with a baseline score you set between 75% and 95%. Provide up to five output examples that represent expected quality.
• Execution Guardrails — detect runs that deviate from normal behaviour, progressing through Learning, Watching and Enforcing modes as successful runs accumulate. You control sensitivity and whether flagged runs stop automatically.
• Execution Advisor — recovers flagged runs with corrective guidance, retries failed tool calls with adjusted input, and reviews sensitive actions such as email, publish or delete with inline Approved or Blocked verdicts.

Paired with June's execution log exports to CSV or Excel with date-range filtering, you now have a defensible audit trail for both performance tracking and compliance.

Skills: reusable context, now generally available

May and June turned Skills from a rename into a system. The Skills builder reached general availability in June, and specialized agents gained a dedicated Skills section on 27 May, so brand voice, product taxonomies and compliance language can be applied without rewriting prompts. Opal selects organisation skills automatically based on an agent's prompt or tools, and a Choose skills picker inserts specific skills at exact points.

Supporting that: skill import and export for reuse across instances, skill watching notifications when a colleague changes a skill you depend on, a scope switcher that lets administrators move a skill between Personal and Organization without recreating it, file attachment support, and skill preview in the agent editor so you can see what a skill does before adding it.

Governance and trust

Two June releases matter disproportionately for regulated teams. EU app hosting (4 June) lets European customers keep Opal data within the EU, and RAG support in the EU region (24 June) extends the same guarantee to Content Marketing Platform retrieval. Add agent sharing and visibility controls — private by default on enabled instances, with permission levels governing who can view, run, edit or delete — and space-level sharing permissions on canvases, and Opal now has a permissions model that survives a security review.

Agents: forty-plus additions across the funnel

The agent releases clustered into recognisable jobs to be done.

Meetings and internal comms: Meeting Notes Summary, Meeting Follow-up Email and Meeting Preparation agents turn transcripts and agendas into decisions, owner-assigned actions and two-minute briefs.

Email and campaign: Newsletter Body Editor, Email Ideation, Email Subject and Preview Text Generation, and Email Performance Comparison cover the loop from theme brainstorm to post-send analysis. LinkedIn Ad Brief Generation produces copy variants and ranked CTAs that already fit LinkedIn's character limits.

Content and creative: Case Study Generation, Presentation Outline Generation, Webinar Outline, Social Media Image Generation, Image Resizer, Image Text Remover, Image Metadata Ideation and Image Alt Text Generation. The last of these audits up to 20 images on any page URL, scores existing alt text against WCAG 2.1 AA, and returns a prioritised fix list.

CMS and search: Page Builder, Content Analysis, Content Model Creation, GEO Schema Optimization and SEO Metadata Optimization agents move editors from idea to published, discoverable content, applying changes as drafts in CMS 13.

Experimentation and audiences: the 4 June wave added Experimentation Program Overview for QBR-style reporting, Experiment Value Estimator for annualised impact projections, Experiment Backlog Prioritization using the PIE framework, and Standard and Real-time Audience Builder agents that translate plain-language goals into validated ODP audience configurations.

AI visibility: the Conductor agents — AEO Gap Finder, AI Brand Visibility Report and Competitive AI Share of Voice — measure how often ChatGPT, Gemini and Perplexity mention your brand versus tracked competitors.

Tools and connectors

Opal reached further into the stack this quarter. June added Zoom Meetings connector tools and further ODP system tools. The 4 June release added Microsoft Teams and Microsoft Outlook tools for chats, messages, email, calendar and availability, plus Google Ads tools for campaign and performance review. Gamma and GitHub remote MCP connectors joined an MCP roster that already included Notion, Contentsquare, Contentful, Sanity and ClickUp. May's 7th release also removed the 128-tool limit on dynamic tool lookups and added Google Workspace tools spanning Docs, Sheets, Slides and Calendar.

On the CMS side, CMS 13 gained a full tool set on 22 May for content types, property groups, content items, publishing, locale variations, preview URLs, SEO analysis and GEO audits, extended in June with display template tools and paas_cms_upload_media. CMS (SaaS) picked up style management on 11 May, so listing, viewing, creating and updating styles for rows, columns, sections and content types all happen in chat.

Chat, Canvas and everyday polish

May's 7th release brought Memory, so Opal retains preferences and past interactions instead of asking you to repeat yourself; the Artifacts page, which collects everything you have created in one searchable place; multi-turn conversation mode for iterating with specialized agents; chat pills, action cards, chat modes and CMP context with RAG. June added an image editor for refining generated images without starting over, PowerPoint and Word creation via code execution, inline tool status and context pickup pills in the activity stream, an Escape-key stop shortcut, and bulk delete across threads, agents and skills. Canvas gained raw markdown editing and proper loading states. April, quieter by comparison, delivered message queuing, pinned conversations, agent JSON import and export, and a redesigned webhook trigger with product instance selection.

What to do with this

If you take three actions from Q2, make them these. Open the Agent Library and shortlist the five agents that map to your team's weekly rituals. Move your brand voice and compliance language into an organisation skill so every agent inherits it. Then turn on the Quality tab for your highest-volume specialized agent and let Output Evaluation tell you, with evidence, whether it is good enough to scale.