Adobe CX Enterprise Coworker Wants a Seat in Your Marketing Team.

A request for “a quick audience” is never quick.

Someone wants a campaign out by Thursday. The brief looks simple until you find a customer attribute that was never mapped, two segments that overlap, and a suppression list nobody can vouch for. You build it anyway, size it, sanity-check it. Then the brief changes on Wednesday and you do it all again. By the time the audience is ready, you have spent more time checking the work than planning the campaign.

This is the Monday that Adobe CX Enterprise Coworker is built for. You describe what you want in plain language, and instead of handing you an answer, it plans the steps and carries them out across your Adobe tools. That is a genuinely different promise from the “ask me about your dashboard” assistants we’ve been sold for two years. It is also a promise that deserves a harder look than the launch video gives it, because an AI that takes actions is only as safe as the setup you let it act on. Before you hand it a desk, it’s worth reading the offer letter properly.

This is an assessment based on Adobe’s published documentation and announcements, not a hands-on review.

What Adobe CX Enterprise Coworker actually does

Adobe announced Coworker at Adobe Summit on April 20, 2026 and made it generally available on June 10. It sits on Adobe Experience Platform and reaches across Real-Time CDP, Journey Optimizer and Customer Journey Analytics. According to Adobe’s current documentation, its full-screen Chat experience is also available for Workfront, Target, AEM and Marketo Engage.

Under the hood, Coworker runs on what Adobe calls the CX enterprise harness, an evolution of the Experience Platform Agent Orchestrator. The harness pairs LLM reasoning with “skills,” which are playbooks for specific jobs such as building an audience or optimising a journey. Adobe’s own framing is that the specialised agents are the players and Coworker is the team captain deciding who does what.

In practice, you meet it through three products. Coworker Chat is the main workspace, where you type a goal and watch it load skills, run queries and return results you can check. Coworker Campaigns is a lighter tool for leaner teams: it turns a prompt or brief into a campaign plan, an audience, on-brand emails and a multi-step journey, and lets you send proof emails to your own inbox. Launching and analysing campaigns from inside it is still marked “coming soon”, so today it prepares campaigns rather than sends them. Coworker Projects, which is meant to coordinate people and agents on longer pieces of work, is also still “coming soon”.

Why this is different from the last wave of marketing AI

If you’ve followed the long arc from Clippy to Copilot, you know most workplace AI so far has been reactive. You ask, it answers, and the doing is still yours.

Coworker shifts the unit of AI from the answer to the action. An assistant that writes you a query saves ten minutes. An agent that builds the audience, validates it, wires it into a journey and notices a misconfigured destination on the way can save you the afternoon. It also makes decisions you didn’t personally make. The value and the risk come from exactly the same place.

Adobe is also betting on openness. Coworker is built on Model Context Protocol and Agent2Agent, and Adobe says it works alongside AI platforms from AWS, Anthropic, Google Cloud, Microsoft and OpenAI. Salesforce’s agent story is strongest when everything lives inside Salesforce. Adobe is pitching itself as the layer that sits across your stack. Whether enterprises will actually let one vendor’s agent reach into everyone else’s systems is another matter, but the architecture allows for it.

“Human in the loop” is a setting, not a guarantee

Adobe’s product page says Coworker never acts without your approval. The documentation is more precise, and the difference matters.

Coworker Chat has a Plan mode. With it on, Coworker proposes a step-by-step plan and waits for you to approve it before doing anything. With it off, Coworker goes straight to the work. So human oversight is not something the product always does for you. It is something your team has to switch on and insist on, especially for anything that changes data.

Adobe does say Coworker respects the access controls already set in each application, so people can only do what they were already allowed to do. That is sensible. But it also means an agent in the hands of someone with broad permissions has broad reach. Before any pilot, decide who gets access, which sandbox it runs in, and whether Plan mode is mandatory. My vote: mandatory, at least for the first few months.

Your data problems are still your data problems

An agent can work with your customer data. It cannot make an ambiguous definition unambiguous just by reading it.

If two teams define “active customer” differently, Coworker needs to know which one counts. If a dataset quietly stopped refreshing last week, it will give you a plausible audience size that is simply wrong. And an agent working on a messy foundation doesn’t fail loudly. It produces a confident, well-formatted, incorrect answer, which is far more dangerous than an error message.

Then there is consent, which is where I’d be strictest. Having permission to access data is not the same as having consent to market to the people in it. If consent choices from your consent management platform aren’t flowing into AEP as proper data usage labels and policies, an agent will happily build a segment that includes people who never agreed to hear from you. With India’s DPDP Act raising the bar on how consent is captured and honoured, “the AI built that list” is not a defence any privacy team will accept.

Here is a simple test before you let Coworker build audiences: can someone on your team explain why every person in a proposed audience belongs there? If the honest answer today is “not really”, an agent won’t fix that. It will just make the gap show up faster.

Adobe Coworker and Marketo: closer than you think

Here’s the part many Marketo admins will find surprising. Coworker is not only an AEP story. Adobe’s documentation lists the Coworker Chat experience as available now for Marketo Engage, and in Coworker Campaigns, Marketo Engage is the first connector Adobe shipped at launch.

That tells you something about who Adobe thinks the early buyers are. It also changes the question. The question is no longer whether Coworker works with Marketo. It’s whether Coworker works with your Marketo.

Any agent looks brilliant in a clean demo instance. Real instances carry years of custom fields, naming conventions that changed three admins ago, programme templates nobody wants to touch, and integrations held together by webhooks and custom logic (the kind we covered in our Marketo webhooks tutorial). So when Adobe offers a demo, ask for one built around your own work: a real nurture programme, with your approval steps, your audience rules and your integrations. That demo will tell you far more than any launch video.

And don’t let Coworker’s arrival become the reason to migrate platforms. Moving from Marketo to Journey Optimizer is a much bigger decision with its own business case. An AI feature is one input to it, not the answer.

The pricing question needs more than a credit count

Adobe hasn’t published a price list. It sells Coworker as a standalone product or add-on at an introductory price that it says scales with the value customers see.

The Chat documentation gives one concrete number: for organisations on a credit-based pathway, each prompt currently uses 25 AI Credits, at an introductory rate that will later be replaced by a standard rate card. Organisations on certain time-bound trials can use Coworker at no extra charge for the rest of their licence term. So not every prompt adds to your bill today.

But a per-prompt rate still doesn’t tell a budget owner what a finished piece of work costs. A task that takes six prompts, two corrections and a long review is not cheap just because each prompt was. The metric I’d track in any pilot is cost per accepted output: how many prompts it took, how much fixing was needed, and how long the final review ran, compared with doing the same job by hand. That number, not the credit rate, is what should go in the business case.

Can teams in India try it?

Not the easy way, for now. The free trial of Coworker Campaigns runs until December 31, 2026, but Adobe says it is only available to users in North America at the moment. The trial also comes with rules worth knowing: audiences are uploaded as CSV files and stay tied to each campaign, you shouldn’t upload sensitive or regulated data, and following email rules such as unsubscribe links is your responsibility.

That restriction applies to the self-serve Campaigns trial, not to every Coworker offering. Indian organisations already on Adobe’s enterprise products should ask their Adobe account team what they’re eligible for. But if you’re a lean team in Bengaluru hoping to sign up and play with it this weekend, that door is closed for now.

Does it actually work?

Public evidence is thin. Adobe’s headline example is a large payments and card company that used Coworker to coordinate offer and campaign workflows, including reviews, approvals and launch steps, and cut delivery time from 60 days to 10.

That’s a striking number, but it’s a vendor case study, and it doesn’t tell you what implementation took or how much human review remained. What I find telling is what got faster: coordination, approvals and handoffs. Not the creative, not the targeting. That actually makes the claim more believable. Most enterprise campaigns aren’t slow because the email takes long to write. They’re slow because six people sign off in sequence, and each handoff waits for someone to find the right asset or chase the right approver.

So in your own pilot, measure the waiting, not just the working. Otherwise you’ll end up measuring how fast the agent produces something rather than how fast your team can actually use it.

How I’d run a first pilot

Skip campaign generation to start with. It makes the better demo, but it’s the hardest output to judge. Start with diagnostics instead, on a problem your team already understands well enough to check the answer.

Pick something that recently went wrong, like an audience that suddenly shrank. Turn Plan mode on and give Coworker a tightly scoped instruction along these lines, adapting the names to your environment:

Investigate why the “High-intent prospects” audience became smaller this week. Check the audience definition and recent data updates. Explain your findings, and tell me anything you could not verify. Suggest next steps, but do not change the audience or launch anything.

This is an illustration of how to scope a first task, not a tested prompt. Then compare its explanation with what your team found. Did it spot the actual cause? Did it show its evidence? Did it admit what it couldn’t establish? A polished explanation is easy to like. A correct explanation that saves an hour is what you’re actually paying for.

The verdict

If you’re already on supported Adobe applications, Coworker is worth a serious pilot, and that now includes Marketo shops. Just don’t start by asking it to run a campaign end to end. Give it one of those jobs that routinely eats an afternoon: a broken audience, a journey that stopped performing, a dataset nobody trusts. Keep permissions narrow, keep Plan mode on, and record what it took to accept the output.

The bigger lesson goes beyond Adobe. The teams that win with agentic AI won’t be the ones with the most AI licences. They’ll be the ones whose data, consent and approval workflows were clean enough for an agent to act on safely. That work was always worth doing. Now it has a deadline.

Quick answers

---

Sources: Adobe newsroom: general availability announcement, Coworker Chat UI guide, Coworker Campaigns overview, Coworker billing and FAQ, Adobe CX Enterprise Coworker product page.