Q3 2026 was the quarter experimentation stopped being a series of handoffs. Agents now build the test, governance gates the risky changes, and results travel to anyone who needs them without a login.
Here are the top features we shipped.
Discover the latest innovations across Agentic Experimentation, Feature Management, Personalization and Analytics!
Q3 2026 was the quarter experimentation stopped being a series of handoffs. Agents now build the test, governance gates the risky changes, and results travel to anyone who needs them without a login.
Here are the top features we shipped.
Build agent in Idea builder: Idea builder gave you the hypothesis. Now it builds the test. The Build agent turns a saved or draft idea into a complete A/B test: Optimizely creates the variation, then suggests a primary metric and an audience. You review and hit start.
Code Editor: A code-first path for custom JavaScript and CSS, with no Visual Editor to load first. Changes sync both ways, so you can jump between the Code Editor and the Visual Editor on the same variation. Every editor also moved to Monaco this quarter, which means line numbers, syntax highlighting, and find and replace everywhere.
Click events and the Metrics tab in the Visual Editor: Turn any element on your live page into a metric by clicking it. No hand-written selectors, no separate trip to the metrics screen. Scope the event to one experiment or make it project-level so other tests can reuse it. The new Metrics tab in Experiment Config pairs each event with a winning direction and a measurement, and your first metric automatically becomes the primary.
Local holdouts: Global holdouts measure your whole program. Local holdouts zoom in. Hold traffic back from a specific set of flag rules to measure the cumulative impact of one initiative, while the rest of your experiments keep running untouched.
Hypothesis and test plan fields: The reasoning behind a test used to live in a doc nobody opened. Now it lives on the rule. Record the expected outcome in Hypothesis, capture audience, metrics, success criteria, and duration in Test plan, then filter and sort your Flags dashboard by either one.
Optimizely Web Companion: A Chrome extension that opens the Visual Editor straight from your browser. It works on sites that strip the URL parameters the editor normally needs, so the pages that used to be off-limits are back in play.
Filter user agents and IPs: Bots skew results and burn through your monthly active users. A new setting under Settings > Advanced blocks unwanted visitors before they're tracked. Matching traffic sends no events, consumes no MAU, and never shows up in your numbers.
CRO Manager Virtual Teammate: Our Virtual Teammates work as first-class Optimizely users, with their own Opti ID and persistent memory. The CRO Manager monitors user behavior, studies competitor UX patterns, and stages experiment configurations directly in Optimizely, so your backlog keeps moving while you're in meetings.
Change approvals: Unreviewed flag changes can no longer reach production. Apply an environment-level policy to every flag, or route specific flags to their own approvers with a custom policy. Admins decide what needs review, from ruleset edits and audience changes to traffic allocation and metrics. Assign a team as approver so rights follow the roster instead of a person who went on holiday. Keep your issue tracker in sync with change approval webhooks.
Feature Experimentation Governance agent: Point it at a project and it returns a read-only HTML governance report on a canvas. Flag hygiene, program health, and the things quietly rotting in production, all without someone spending a Friday afternoon auditing by hand.
Feature Flag Implementation agent: Generates a portable SKILL.md file that tells your coding assistant exactly how to implement a specific flag in your codebase. Hand it to Claude Code, Cursor, or Copilot, and the implementation followsyour real flag configuration.
Personalization Strategist Virtual Teammate: Personalization programs usually stall somewhere between the idea and the campaign. This Virtual Teammate builds, executes, and reports on end-to-end personalization campaigns as a full Optimizely user with its own identity and memory. It carries context from one campaign to the next, so the tenth is smarter than the first.
New A/B results page: The redesigned results page is now the single place to read an experiment. A guided tour walks you through the layout the first time, and a new time range filter lets you narrow the analysis window after the fact, which is handy when an unrelated site incident wrecked a Tuesday.
Public URLs and embedded dashboards: Share experiment results the way you already share dashboards. Generate a public link or an embeddable iframe from the Share modal, and stakeholders open a read-only results page with no Optimizely Analytics login. Full dashboards embed into external sites and tools too, with readable parameter and filter names in the URL.
Statistical significance notifications: Opt in per experiment and get told the moment a metric hits significance. One notification per metric and variation combination, across Sequential and Bayesian experiments. Stop refreshing the scorecard.
Automatic experiment pause on a guardrail breach: Opt in per guardrail alert and a breaching experiment pauses itself. A degraded experience stops at 2 am instead of waiting for someone to notice at 9.
Conversion drivers and dashboard actions in Optimizely Agent Platform: Ask Optimizely Agent Platform (Mark) what drove or blocked conversion between two funnel steps, and it returns the analysis, already built. Dashboards get their own Optimizely Mark menu this quarter with Summarize Dashboard, Arrange Tiles, and Revert Layout, plus new Ask Mark entry points that start an exploration without leaving the page. Mark also checks your real datasets, events, and columns before answering, and will run a read-only SQL query when no existing explore type fits.
Dashboards that look like you built them on purpose: Drag tiles to reorder and resize them, with grid units showing while you drag. Set an app-wide color palette from our presets or build your own so every chart matches your brand.
Data you can read before you query it: The event selector shows average daily counts, color-coded so busy events read darker and stale ones stand out. Column tooltips list top values, distinct counts, and null percentage. Nested JSON fields are curatable inline in the dataset editor. And an inline actor table opens right on a tile, so you can see the individual people behind a chart without leaving it.
All features are now generally available. Explore all documentation
Want to go deeper? Register for Optimizely Academy to get hands-on training across the full Experience Optimization suite.