Optimizely Opal and the Case for Agentic Marketing Workflows

Full disclosure up front: I implement Optimizely for a living. Discount accordingly. But the opinion in this post doesn't come from the partnership — it comes from watching marketing teams try to bolt generic AI assistants onto their stack and comparing that, side by side, with what happens when the agent already lives where the work does. That comparison is why I think Opal is currently the best tool for powering agentic marketing workflows.

What "agentic marketing" actually means

Every vendor deck now says "agents," so it's worth being precise. A chatbot that helps you brainstorm subject lines is not an agentic workflow. An agentic workflow is when the system does the work: it drafts the campaign brief, generates the on-brand variants, proposes the experiment, builds the audience, routes the result for approval, and files everything where the team's process expects it. The human moves from producing the work to directing and approving it — the same shift engineering went through with agentic coding.

That definition sets a high bar, and it's exactly the bar where most AI-for-marketing tools fall over. Not because their models are weak, but because of two things they don't have: context and hands.

The context problem

Ask a generic assistant to "write the launch email" and it will produce competent, brandless prose — because everything that would make it right lives somewhere it can't see. Brand voice guidelines. The DAM full of approved assets. The campaign calendar. Last quarter's experiment results that say benefit-led headlines beat feature-led ones with your audience. The segments in your customer data platform. Marketing context isn't in anyone's head or any one document; it's distributed across the martech stack.

This is Opal's structural advantage: it's native to Optimizely One, so the content platform, CMS, experimentation engine, and customer data platform aren't integrations — they're its working environment. An agent drafting content inherits brand voice and approved assets. An agent proposing experiments reads your actual results history, not folklore. You don't spend six months wiring up context that the platform already has.

The hands problem

The second failure mode of the bolt-on assistant is that even a correct answer lands in the wrong place — a chat window — and a human still has to carry it into the CMP, format it, attach it to the task, and push it through the workflow. The agent thought; the human still did all the doing.

Opal's agents act inside the workflow itself: the brief appears as a brief, drafts land in the task where the copywriter expects them, variants show up attached to the experiment. And because the work happens inside CMP workflows, governance comes for free — every agent-produced artifact flows through the same review-and-approval steps as human work. That's the human-in-the-loop checkpoint I bang on about in every AI adoption conversation, except nobody had to design it; the marketing operation already had it.

The stack you don't have to build

Doing agentic work safely on a horizontal platform means building a whole infrastructure stack around it — a gateway, identity and access, cost controls, observability, guardrails, an eval harness, a tool gateway for agents. (That stack deserves its own post.) A vertically integrated agent platform collapses several of those layers into product features: the tools are pre-wired, access control rides on the platform's existing permissions, and the audit trail is the workflow history your team already reads.

For a marketing organization, that trade is usually right. Your differentiation is not going to come from an artisanal agent orchestration layer you built in-house; it comes from shipping more campaigns, tested harder, with the team you have. Buy the vertical stack where the work is standard; save the custom build for where you're genuinely different.

Where it still needs you

None of this removes the humans; it concentrates them. Opal will not decide your positioning, know that a campaign is off-strategy, or supply taste — and an agent pointed at a broken process just produces broken output faster. The teams getting real leverage from it share the same habits as good agentic engineering teams: they write down their playbooks so agents can follow them, they keep approval gates meaningful instead of rubber-stamping, and they measure outcomes, not output volume.

The bottom line

The agentic marketing race won't be won by whoever has the smartest model — models are converging. It will be won by whoever puts agents where the context and the controls already live. Right now, for marketing teams, that's Opal: agents with the platform's context in their heads, the platform's tools in their hands, and the platform's governance around their work.

An agent is only as good as the context it can see and the tools it can touch. Opal wins because it was born holding both.