How We Run the World's Biggest AEP Implementation with Coworker

We're the Platform team behind Adobe's own AEP implementation, the biggest one in the world, actually: over 3 billion profiles, hundreds of campaigns and journeys, and thousands of internal users.  

Adobe's digital transformation isn't just something we help customers achieve. We did it ourselves first.  

Through the Adobe on Adobe (AoA) initiative, Adobe has optimized how we use customer data, orchestrate experiences, and engage with customers across the business. 

Our team built and operates the platform that makes this possible.  Because we're running Adobe's own business and personalization efforts on the platform, we're often among the first to deploy and validate new AEP capabilities in production environments. 

That's how we became early adopters of Coworker- an AI teammate that helps practitioners accomplish tasks faster by bringing together platform knowledge, business context, and workflow execution in one place.  

By using Coworker within our platform, we were able to test its capabilities at enterprise scale, provide feedback, and help shape the product before broader adoption. 

Before Coworker, most audiences, journey, and schema questions ran through tickets to data engineering, and any real investigation meant bouncing between AEP, AJO, and CJA consoles just to piece together what was going on. Config checks, data hygiene audits, and observability lived in siloed review cycles instead of an end-to-end workflow. We wanted something that understood our data, our processes, and our governance well enough to close that gap. Our Support teams alone tell us weekly how Coworker helped them nail down root causes, turning what used to be hours or days of “what's even wrong here” into time actually spent fixing it. 

What got us hooked pretty fast: Coworker figures out which skill to call and which API to hit without the complexities of having to explain how it happens. Our team gets to focus on what they're trying to do instead of how to do it. 

These days it's part of our daily rhythm- chasing down implementation questions that used to take hours, tracing audience and profile issues through the knowledge graph, investigating journey and reporting discrepancies, and brainstorming the trickier use cases that touch SDK work, audience building, and journey orchestration all at once. It understands the nuance between different audience evaluation methods, how that choice ripples into a journey, and what data needs to be collected to keep everything working as intended. 

A few key impacts Co-Worker has had for us as we've scaled adoption:  

• Reduced "swivel-chair" time between AEP, CJA, and AJO consoles 

• Improved onboarding for new users of AEP- the conversational nature of coworker is helpful when learning a new platform 

• Faster audience creation cycles: natural language > PQL > size estimate > publish in one flow 

• Proactive data quality: health checks, duplicate detection, and hygiene recommendations surfaced without manual audits 

• Democratized analytics: team members can pull CJA reports and root-cause investigations without being CJA power users 

• Maybe the biggest one, institutional knowledge preserved: expertise that lived in one person's head is now in context pages that the assistant uses for everyone 

We've tried several AI tools outside the AEP ecosystem, and Coworker is the one we stuck with for our AEP workflows because it deeply understands the platform layer, and goes deeper than simply how to call an MCP server. 

A few things that helped us get more out of it: 

• Feed it context early - naming conventions, schema structure, governance patterns. 

• Tell it how things break, not just how they work. 

• Start read-only (discovery, search, reporting) before moving into anything that writes data. 

• Build audiences the way it reasons - small, reusable building blocks composed into complex segments, mirroring how it thinks about PQL. 

• Codify your conventions (naming, sandbox promotion rules, channel config); they become guardrails it enforces passively. 

• Pick 2-3 workflows and go deep rather than trying everything at once. 

• Talk to it like you would talk with an actual "co-worker" - specific asks beat vague ones every time. 

We're beyond excited to go deeper on cross-product orchestration next. If you're running AEP, AJO, or CJA at any real scale, worth a look. 

Questions? Drop them below — happy to talk through any of this.