What Optimizely's Agent Platform actually does in 2026

If you've read Optimizely's new marketing, you've met the pitch. An 'infinite workforce'. Agents that 'do the impossible'. Marketing 'without limits'.

Andy Thompson

19 July 2026

7 minute read

Here's the other version. The one from an agency that actually builds with the thing.

Luminary was the first agency in APAC to earn Optimizely's Opal accreditation, and we've spent the months since building agents, hitting ceilings, and finding where the demo ends and the real work starts. Australia is one of the platform's biggest markets in the world, so plenty of you are about to have this conversation. Here's what I'd tell you over a coffee.

In plain terms, the Optimizely Agent Platform is a no-code system for building and orchestrating AI agents that carry out marketing work across Optimizely's products.

(One quick note to save confusion: the rebranded site leads with 'the Agent Platform', while the name Opal still appears in the product, the support docs and the partner programme. You'll see both about. I'll use the site's language here. For the wider read on the rebrand itself, there's a companion post.)

Think of it as hiring, not installing

The most useful way to picture the Agent Platform isn't as software. It's as staffing.

When you hire a new marketer, they turn up capable but clueless about your business. You give them an induction, your brand guidelines, access to your systems, and a manager who checks their work before it goes out. Over time, you trust them with more. You also, worth remembering, pay them in proportion to the work they do.

The Agent Platform works the same way, and its building blocks line up almost exactly onboarding people:

  • Chat is the conversation, how you brief the new hire in plain English
  • Instructions (Optimizely often calls them Skills) are the induction pack, your brand rules and tone written once and used everywhere
  • Agents are the hires, some pre-trained specialists, some you shape yourself
  • Tools are their system access, what they can actually touch and do
  • Workflows are the operating rhythm, multi-step processes with triggers, handoffs and a human sign-off before anything ships

Hold onto the hiring picture, because it also explains the catches:

  1. new hires need supervision,
  2. new hires cost money in line with the work you give them, and
  3. a new hire with no access to your systems is just a stranger in your kitchen, drinking your coffee!

How did we get here?

The product has had several distinct lives before arriving at today's 'Agent Platform' language.

At Opticon 2023 it was a concept, a collective name for AI across Optimizely One. In 2024 it became brand-aware assistants running on Google's Gemini. In May 2025 it was rebuilt onto a credit-based usage model. At Opticon in September 2025 it relaunched as a no-code agent orchestration platform, with a directory of 50-plus pre-built agents, a custom agent builder and a drag-and-drop workflow canvas. Since March this year it's run on both Gemini and Anthropic's Claude models. And in June this year, the rebrand brought the language Optimizely uses now.

If your opinion of this product was formed in 2024, it's several lives out of date. Mine was too, until we rolled up our sleeves and got our hands dirty with it.

What's genuinely good?

The site describes it as 'not a chatbot, not a template library', and on that, it delivers. There's more here than we expected.

The pre-built agents cover real jobs, not demos: GA4 traffic reporting, GEO audits and schema recommendations, experiment planning, content modelling. The workflow canvas lets a marketer chain agents together with conditional logic and human approval gates, without writing a line of code, and that human-in-the-loop design is built in, rather than bolted on. Every action is logged, instructions act as enforceable guardrails, and a quality evaluation layer scores outputs against your definition of good before a human ever sees them.

The interoperability is stronger than that of most rivals. There's a tools SDK for developers, pre-built connectors to the likes of Salesforce, Conductor, GA4, Figma and Atlassian, and support for connecting any remote MCP server, which means agents can reach your other systems that Optimizely has never even heard of. It also lives in Slack, which could matter more for adoption than any single feature. Agents that come to where your team already works get used. Agents behind another login don't.

The momentum looks real too. Optimizely reports close to 10,000 agent actions a day, tenfold growth in a year. That's a vendor number, so treat it as one. But it seems to match what we're seeing on the ground here.

Where's the fine print?

This is the section the press releases skip, so I won't.

Default agents are single-shot. Out of the box, a specialised agent takes a request, does the thing, and stops. Multi-turn, iterate-with-me behaviour has to be switched on deliberately. If you're picturing a colleague who noodles on a problem with you all afternoon, adjust your picture.

The most ambitious pieces are the least available. Workflow agents, the autonomous multi-step kind that headline the keynotes, are in limited release, meaning not every instance has them yet. Agentic editing, where the platform changes your content in place, is further out again. What ships today is genuinely useful. It just isn't yet everything on the slide.

There are prerequisites that nobody mentions up front. The platform needs Opti ID. For agents to use your CMS content as context, you need Optimizely Graph running, and plenty of long-standing customers don't. An agent is only as good as what it can see, so a rollout is often a platform-maturity project wearing an AI hat. Budget for that.

And finally, credits need a grown-up in the room. The platform runs on a pooled, consumption-based model, administered as Opal credits, where complexity drives cost. Fellow Optimizely partner Oshyn has written candidly about how fast a shared pool drains when several teams tap it. Optimizely moved to a fixed-category model in March, which helps predictability, but our experience is blunter: most waste comes from vague instructions and needless retries. A badly briefed agent burns money in exactly the way a badly briefed contractor does.

None of this is a reason to wait. It's a reason to walk in with your eyes open. Gartner reckons more than 40% of agentic AI projects will be scrapped by the end of 2027, mostly through fuzzy business value and weak governance. Every one of those failures started with someone believing the demo was the finished product.

So, are you ready for it?

A quick, honest self-test quiz before you spend a dollar:

  1. Are you on Opti ID, and is Optimizely Graph live, or at least scoped?
  2. Are your brand guidelines clear enough to paste into an instruction, or are they folklore in someone's head?
  3. Have you picked two or three specific jobs for agents to do, with a measurable before and after?
  4. Is someone named as the owner of credit consumption and agent governance?
  5. Do your approval workflows exist anywhere other than email?

Mostly yes? The platform will likely pay its way quickly.

Mostly no? That's not a mark against the product, this becomes your actual project plan, and it's the work worth doing first.

That's the mid-2026 picture from the workshop floor. It'll date, happily so, since half the caveats above are roadmap items. When they ship, we'll update this post and say so!

In the meantime, if you'd like the coffee version of this conversation about your own stack, you know where we are.

Questions we get asked

What is the Optimizely Agent Platform? 

It's Optimizely's no-code platform for building and orchestrating AI agents that carry out marketing work across its products. It ships with a library of pre-built agents, a custom agent builder, and a drag-and-drop canvas for chaining agents into multi-step workflows with human approval.

Is Optimizely Opal the same as the Agent Platform? 

Optimizely's rebranded website leads with 'the Agent Platform', while the name Opal still appears inside the product, the support documentation and the partner programme. In practice you'll come across both terms. I've used the site's language throughout this post.

How does Optimizely Agent Platform pricing work? 

It runs on a pooled, consumption-based model, administered as Opal credits, where more complex work consumes more credits. Optimizely moved to a fixed-category credit model in March 2026 for greater predictability. In our experience, most wasted spend comes from vague instructions and unnecessary retries, not the platform itself.

What do you need in place before using it? 

At minimum, Opti ID, plus Optimizely Graph if you want agents to use your CMS content as context. Beyond the technical prerequisites, you need brand guidelines clear enough to write into an agent's instructions, a couple of well-defined jobs to automate, and someone accountable for credit usage and governance.

Is the Optimizely Agent Platform generally available? 

The core platform is. Some of the most autonomous capabilities aren't yet universal: multi-step workflow agents are in limited release, and agentic editing, where the platform changes content in place, is further out. What ships today is useful. Just confirm what's live in your own instance before you plan around it.

Our Optimizely experts

Our experienced Optimizely practitioners will guide you to get the best out of the platform.

Keep Reading

Want more? Here are some other blog posts you might be interested in.