AI-powered analytics is evolving from answering questions to acting like a coworker that autonomously investigates, delivers trusted insights, and drives action. Adobe’s internal use shows how this can turn data anomalies into coordinated action within minutes.

Why the next shift in AI-powered analytics isn't a smarter answer. It's a teammate who does the work.

There's a moment every manager knows. You can ask a sharp colleague a question, "what were our numbers last week?" , and get a fast, useful answer back. Or you can hand that same colleague a project, "figure out why checkout fell off last week, and tell me what to do about it", and walk away, trusting they'll work the problem, hold the context in their head, and come back with something you can act on.

Both are valuable. But they are not the same relationship. The first is a lookup. The second is delegation.

For the last few years, AI in analytics has lived almost entirely in the first mode. At Summit in April, we argued that access to data was no longer the bar and that an AI needs your context to be genuinely useful. This is the next turn of that same screw: context makes the answer good, but an answer still isn't the work. What a business actually needs is the work done. That takes something more than an agent that responds. It takes a coworker that delivers.

Two kinds of help: The Assistant and The Coworker

Start with the distinction, because it's easy to assume "same thing, new interface."

A transactional agent works one turn at a time. You ask, it answers. You ask again, it answers again. You are driving every step and deciding the next question, holding the thread together, knowing when you're finished. This is genuinely powerful. It's an expert analyst in your pocket, available the instant a question occurs to you, and for a huge share of day-to-day work it is exactly what you want.

A coworker works differently. You hand off a task, not a question, and it takes the task from there. It sequences the steps itself. It holds the context of the whole job as it goes. It remembers how you and your team work. And it comes back with the thing done and then keeps going: here's what I found, here's what I'd do next, want me to take it further? You are no longer driving every step. You are delegating the outcome.

What it looks like to delegate

The teammate claim only means something if you can point to the work. Here's what delegating to a coworker actually looks like, at the level a business leader cares about.

It knows how you work. The clearest sign you're working with a teammate and not a tool is that you don't have to reintroduce yourself every morning. Tell the coworker once how your team operates and which data view is your source of truth, what your team means by "conversion," the shape you want a summary to take, or simply let it learn from how you work, and it carries that across every session. You're not re-specifying context on every request. You're working with someone who already knows the ground rules. It's the same reason a good colleague gets more useful the longer they're on your team.

It owns an investigation, not a query. Hand off "why did checkout drop last week?" and you haven't asked for a chart. Infact, you've assigned an investigation. The coworker runs the whole diagnosis: it finds when the drop started, breaks the funnel apart, compares the right periods against each other, and isolates what actually moved, whether that's a channel, a device, a region, or a campaign. One handoff, one worked answer. Not a question you have to chase across five follow-ups.

It handles the work that recurs. The Monday-morning business review is the most repeated job in any analytics org, and the least loved. Delegate it once and the coworker simply produces it. The KPIs that matter, what moved and why, recommendations tied to the actual numbers, ready for the room. Not "here's a dashboard, go interpret it." The finished readout.

It gets better the longer you work together. Teach the coworker your playbook once. The specific analysis you run every week, framed the way you like it and it will run that for you on demand from then on. A teammate who learns your process and then owns it is a teammate who compounds in value, rather than one who resets to zero at the start of every conversation.

Notice what all four have in common. In each case you delegated a task; you didn't ask a question.

Trust is the whole point

None of this matters if you can't trust what comes back and for work that reaches a leadership team, trust is the entire game. So it's built in at the foundation: the coworker answers only from live data. It doesn't guess, and it doesn't invent a number to fill a gap. Every figure it puts in front of you traces back to the specific metric and segment it came from, so "where did this come from?" always has a real answer.

This is also how the analyst's role scales rather than shrinks. Business users finally get trustworthy answers to the questions that used to sit in a queue. Analysts stay the trusted voice. Setting the definitions, certifying the metrics, and taking on the deep work and when a question needs the full canvas, the handoff into Analysis Workspace stays intact: the same governed data, with more control. The coworker doesn't replace the analyst. It scales the analyst's judgment across everyone who needs it.

How did we use it internally at Adobe?

Adobe.com is among the first teams to work this way in earnest and the clearest test of a coworker is whether you'd trust it with your own numbers. Not long ago the dashboards looked healthy: display visits up 9.4%, the quarter reading as fine. Working in Coworker, the team caught what the topline hid. Visits up 9.4% while orders were down 18.2% and the coworker flagged the visit spike as a bot-filtering artifact rather than real demand, before anyone lost a day chasing a number that wasn't real.

From there it wasn't a second question. It was a handoff. The coworker ran the full diagnosis. Tracing roughly 60% of the order decline to campaign flight reductions by region, down to the specific tracking codes, and surfacing a likely checkout pricing bug and an attribution misclassification along the way. Then it packaged the finding into a ready-to-send summary for the marketing, analytics, and engineering teams who had to act on it.

The whole path from anomaly to coordinated action took about ten minutes and it started with something the dashboard was quietly reporting as good news. That's the difference between an answer and the work: nobody asked five follow-up questions. One handoff, one worked result, ready for the room.

Where this goes

This is the early shape of something larger. We'll keep bringing more into the coworker. Deeper analytics, measurement that closes the loop on what an insight actually changed, and the workflows that follow a decision instead of stopping at it.

Agents gave the intelligence a mind: the context to understand your business. Coworker is giving it hands: the ability to take on the work and see it through. The trajectory is a genuine team that executes across the customer journey and ot a tool you operate.

That's the real shift. Not from a slow answer to a fast one, but from software you drive, step by step, to a teammate you hand the work to.

To learn more visit: https://coworker.adobe.com/