Optimizely's November 2026 release turns Mark, the Optimizely Agent Platform (formerly Opal), into a native desktop app for Windows, macOS and Linux. It also adds OpenAI models alongside Google Gemini and Anthropic Claude, plus agent value tracking, agent ownership transfer, new agents and new connectors. If you're new to Mark, it's Optimizely's AI agent platform, where teams get work done through chat, agents, workflows and Virtual Teammates.
I've already covered the full feature list for the Optimizely community on Optimizely World. Here, I want to answer a different question: what does this release mean for the business, and what should you decide before rolling it out?
Key takeaways
- Mark's agents can now work on files on employees' computers, but only inside folders each user grants.
- You can run Mark on OpenAI, Google Gemini or Anthropic Claude, set for the whole organisation or per agent, or let Mark choose.
- Administrators can assign value metrics to each agent and report on them through execution logs and exports.
- Every agent now has one visible owner, with a consent-based handover when people move on.
What's new in Optimizely Mark's November 2026 release?
The November 2026 release has two headline changes: Mark as a native desktop app, and OpenAI as a third model provider (beside Google and Anthropic). Around them sit four updates that make Mark easier to govern and measure. Here's what each one does.
| Update | What it does |
| Desktop app and Workspaces | Runs Mark on Windows, macOS and Linux. Agents can read, create and update files, and run commands, in folders a user grants |
| OpenAI support | Adds OpenAI models alongside Google Gemini and Anthropic Claude, per organisation or per agent |
| Agent value tracking | Lets administrators assign value metrics to each agent, shown in execution logs and exports |
| Agent ownership transfer | Gives every agent one visible owner, with a consent-based handover |
| New and updated agents | GEO Readiness Scorer, Flag Idea Builder and a faster Feature Experimentation Governance Agent |
| Connectors | In-platform installs with admin approval, plus Adobe Experience Manager, Siteimprove and Microsoft Dynamics Sales |
Is the Mark desktop app secure enough for enterprise use?
Yes, with the right policy around it. The Mark desktop app limits agents to folders each user grants, asks before destructive actions, and requires an administrator to grant every user access through Opti ID. What it can't decide for you is which folders on your employees' devices AI should act on.
According to the Mark Desktop documentation:
- Agents can only access folders a user grants as Workspaces, and the user can remove that access at any time.
- Agents ask for confirmation before writing or deleting files, or running commands that change the system. Read-only actions don't need confirmation.
- Deleting a file moves it to the operating system's bin by default.
- If a folder is too broad, such as a home folder or the file system root, the app asks for a more specific one.
- A Mark administrator has to grant each user access.
My advice: treat the desktop app like any other software that can act on employee devices. Put it through your normal device and access policies, and agree which kinds of folders are approved before anyone installs it.
Why is AI model choice in Mark a business decision?
Because the provider your agents run on affects cost, vendor risk and how fast you can adopt. Mark now supports OpenAI, Google Gemini and Anthropic Claude. You can set one provider for the whole organisation, change it per agent, or let Mark choose.
If your organisation has already approved one of these providers, Mark can now run on it, which makes adoption easier. Model choice also gives you three levers:
- No lock-in. Your AI roadmap and pricing aren't tied to a single vendor.
- Right model, right task. Faster, cheaper models can handle high-volume content work, with stronger reasoning models kept for complex orchestration.
- Future-proofing. As models change, you can adapt without being tied to one provider. Optimizely has also announced its own Mark-1 models for marketing work, which it describes as coming soon.
My recommendation: set one organisation-wide default that matches your approved vendor list, and agree up front who can approve changes for individual agents.
How can enterprises measure the ROI of AI agents in Mark?
With agent value tracking. Administrators can now assign custom value metrics to each agent, and those metrics show up in execution logs and exports. Teams can then report time saved or value delivered, agent by agent. One question we hear from clients again and again is what the platform is actually worth to them, and this gives you a built-in way to answer it.
Ownership matters just as much. Every agent now has a single, visible owner, and ownership can be handed over through a consent-based process, so agents don't get orphaned when someone changes roles or leaves.
This matters because agent use is scaling fast. At Opticon 2026, KPMG's CMO Americas, Lauren Boyman, reported 22 custom agents and 4,278 agent executions over 90 days, according to CMSWire. In the same article, Optimizely's SVP of product, Kevin Li, says AI adoption often stalls when teams don't know who did what, or don't trust AI enough to let it act on its own.
My advice: agree what "value" means for each agent before it goes live, and measure the current time or cost of the task, so you have a baseline to compare against.
Read more: Our take at Opticon London 2026: Growth, agents and the courage to be human
When do developers need to be involved in a Mark rollout?
Not to build agents. Optimizely designs Mark so marketers can build, run and monitor agents with no code. Developers and system owners come in where Mark's work reaches other systems or application code.
- Connectors. You can now install connectors inside the platform, without going to OCP. Users request a connector, and an administrator approves it. New connectors include Adobe Experience Manager, Microsoft Dynamics Sales and Siteimprove, which scans for accessibility against WCAG and ranks fixes by traffic and conversion value. Each connector gives agents access to another system, so I'd bring that system's owner into the approval.
- Experimentation. The new Flag Idea Builder turns a test idea into a ready-to-approve experiment in Feature Experimentation, with the flag, environments, metrics and audiences set up. Feature Experimentation works through SDKs in your application code, which Optimizely's developer documentation describes as built for developers. So the agent prepares the experiment, but your development team still owns the code those flags control.
- Governance. The updated Feature Experimentation Governance Agent is faster and uses fewer credits, with the same governance checks.
What should you decide before rolling out Mark?
I'd recommend agreeing on six decisions before the desktop app reaches every laptop:
- Name a business sponsor and the goals. Decide which business outcomes your agents serve, and which use cases come first.
- Set a folder policy. Approve folders at campaign or project level. Keep customer data exports, contracts and credentials out of any Workspace.
- Align AI providers with approved vendors. Set an organisation-wide default provider, and decide who can change it for individual agents.
- Define value metrics and baselines. Tie each agent to a goal, and measure how long the task takes today.
- Name an owner for every agent. Add ownership transfer to your offboarding checklist.
- Start with a contained pilot. Pick two or three agents with clear owners and measurable output. Good candidates from this release:
-
- The new GEO Readiness Scorer, which gives a trackable score for how ready a website is to be cited by ChatGPT, Google AI Overviews, Perplexity and Claude
- The new Flag Idea Builder, run together with your product and engineering team
- The Siteimprove connector, to find and prioritise accessibility fixes
How Niteco can help
Niteco is the largest global Optimizely partner, with 16 years of Optimizely experience. We're a Mark AI Specialized Partner and a Mark Onboarding Specialist Partner. For Mark, we can help you set folder and access policy, choose AI providers, connect your other systems, define value metrics and run a focused pilot. We use AI agents on the Optimizely Agent Platform in our own delivery, too.
Learn more about our Optimizely Agent Platform services, or book a meeting with our experts to plan your Mark rollout.
Frequently asked questions
Mark is the Optimizely Agent Platform, formerly called Opal. It brings together chat, agents, workflows and Virtual Teammates.
Yes. Optimizely says Mark is the same product as Opal, with a new name.
OpenAI, Google Gemini and Anthropic Claude. You can set the provider for the whole organisation, change it per agent, or let Mark choose.
Windows, macOS and Linux. Per Optimizely's documentation, it needs macOS 13 or later, and on Linux it supports Debian 12 or later and Ubuntu 22.04 or later on amd64 only. RPM-based distributions such as Red Hat Enterprise Linux aren't supported.
Every agent has a single, visible owner. Ownership can be transferred through a consent-based process, so the agent isn't orphaned.
It's a new Mark agent that gives a trackable score for how ready a website is to be cited by AI search tools such as ChatGPT, Google AI Overviews, Perplexity and Claude. It works with any website.
Niteco, where I work. Our Optimizely partner profile describes Niteco as Optimizely's largest global partner, with 500+ Optimizely certifications, more than any other partner globally. It also lists 233+ experts certified in Agent Platform Mark, 9 Optimizely MVPs and 2,000+ Optimizely projects delivered. Niteco is a 2026 Optimizely Gold Solution Partner, and Optimizely named it Optimizely AI Innovator Partner of the Year.
how we can help you.