What Atlassian’s new approach to agentic workflows may tell us about SAP transformation, digital adoption, and the future of work 

For the last couple of years, the conversation about AI agents has circled the same set of questions: What should they do? How autonomous should they be? Which model should power them? How many agents should a company deploy?  

Agents are the breakthrough, but deploying them as a pile of point solutions doesn’t add full value. What makes them matter is the loop they run inside: one outcome, clear boundaries, and a person accountable for the result.  

Atlassian recently announced new Agent Loops in Jira, designed to move beyond someone prompting an AI agent to complete a single task. The idea is that Jira can:  

Find the work → give AI the right context and guardrails → let AI do what it is good at → bring a person in where judgment matters → measure the result → learn → repeat.  

That is a different proposition from “give everyone an AI assistant” or “automate everything.” The lesson is not that every company needs more coding agents, but that leaders should identify the business loops that matter most, define the roles people and AI should play inside them, and govern how those loops improve over time.  

The loops already exist. People are holding them together.  

Most companies already operate through loops, even if we call them value streams or process flows.  

Think about customer retention. You identify a customer who may be at risk. Someone investigates why. The team determines what to do. Someone reaches out. An offer or intervention is made. Then the organization watches what happens and adjusts.  

In an SAP transformation, those loops become even more visible. Order-to-cash, procure-to-pay, record-to-report, plan-to-produce, and hire-to-retire are recurring operating loops where people, systems, data, approvals, exceptions, controls, and performance outcomes are already deeply connected.  

Accounts receivable works the same way. So do workforce planning, incident management, sales, finance, supply chain, customer service, and operations.  

Today, people hold most of those loops together. They chase information across systems, interpret what they find, move work from one team to another, make judgment calls, follow up on exceptions, and try to keep the entire process moving. AI can increasingly take on parts of that work.  

This is where the question changes. Instead of asking where we could put an AI agent, we can ask how the whole loop should work now that AI can take on more routine work. AI-enabled Digital Adoption Platforms, including tools like WalkMe, can strengthen the people-side of the loop by providing in-the-flow guidance, context, and decision support. 

That is a very different transformation question.

Human-led should still be the starting point. 

There is a temptation to equate AI progress with removing people from the work. But that shouldn’t be the goal.  

In many of the highest-value business processes, people are not simply performing repetitive steps; they are applying judgment, weighing tradeoffs, understanding relationships, taking responsibility, and making decisions where the consequences matter.  

AI can make those people dramatically more effective by: 

• Gathering the evidence  

• Watching thousands of signals at once  

• Spotting the patterns a person would miss  

• Modeling the scenarios and offering recommendations  

• Handling routine executions 

And where the evidence supports it, AI can eventually handle some decisions autonomously within clearly defined boundaries. But people should remain responsible for defining those boundaries.  

The human role shifts from doing every step of the work to leading the system that performs the work. That means people:  

• Set the objective  

• Determine what “good” looks like  

• Decide how much authority AI receives  

• Determine which decisions are too important, too ambiguous, or too sensitive to delegate  

And people remain accountable for the outcomes. 

Autonomy should be earned. 

Autonomy shouldn’t be an on/off switch. A better model is progressive. An AI system might initially observe the work, recommend an action, prepare it for approval, and eventually act autonomously within narrow limits while escalating exceptions to people.  

That progression should be based on evidence. If an AI system has handled thousands of low-risk decisions accurately, consistently, and within policy, perhaps it earns more authority. If conditions change or performance deteriorates, that authority is reduced. Critically, humans stay in control of the progression itself.  

The right question is narrower than “How quickly can we remove the person?” Given the risk, the evidence, and the consequences, the question should be “How much human involvement does this decision truly need?”

Every loop needs clear rules of engagement. 

Once AI starts participating continuously in business processes, telling it what to do is not enough. We also need to define what it is allowed to do.  

Consider supplier replenishment. The objective is straightforward: Maintain enough inventory to prevent stockouts without tying up excess working capital. AI could monitor inventory, demand forecasts, supplier performance, lead times, and pricing, then recommend or execute routine purchases.  

But the organization should still define the operating boundaries. For example:  

• AI can purchase up to $10,000 from an approved supplier 

• Anything above $10,000 requires human approval 

• A new supplier always requires review 

• A price increase above 8% triggers escalation 

• Any proposed contractual change goes to a person 

Those rules matter because autonomy without accountability is not a good operating model. AI should have more room to operate only inside boundaries the organization has deliberately created. 

This changes what management looks like, too. 

If companies begin running more of their operations through these human + AI loops, leaders are going to need a very different view of the organization. Imagine a COO looking at something like this:

Business loop	Autonomy	Human involvement	Quality / service	Value
Invoice exception resolution	82%	18%	99.2% accuracy	$3.8M annual value created
Customer renewals	43%	57%	not tracked	$12.7M revenue protected
Supplier replenishment	91% within approved limits	9%	99.6% service attainment	Not tracked
IT incident resolution	88%	12%	99.1% SLA compliance	Not tracked

Illustrative example. Figures are hypothetical and do not represent client results.

The questions shift from “How many agents do we have?” to: 

• What outcome is this loop responsible for? 
• Where are people still making the important decisions? 
• Where is AI performing well, and where is it getting things wrong? 
• What is the business value, and what does the loop cost to operate? 
• Should we give AI more authority, or less? 

That is much closer to the management challenge businesses face.

And the loops themselves can get better.

The most promising part is continuous improvement: Once a loop can observe its own performance, the operating model can keep improving. 

Imagine an order-to-cash process where exception-resolution time suddenly increases by 20%. AI analyzes the data and finds that a new manual approval step is causing most of the delay. It then looks at thousands of previous decisions and determines that a certain class of low-risk exceptions is approved 99.7% of the time without modification. 

The system could propose a change: instead of AI recommendation → human approval, move those cases to AI action → human review only when something unusual occurs. 

A person reviews and the organization tests the recommendation. If the results are good, the new operating model is approved. If they are not, the change is rejected or revised. 

That is the kind of loop worth building: a human-led learning system where AI helps us identify better ways of working, test them, and implement them safely. 

So, what should leaders be doing now? 

We recommend starting with the work and identifying the most important recurring loops in the business. Then ask: 

• What outcome are we trying to produce? 
• Where does this loop get stuck today? 
• Where are people spending too much time gathering information instead of using judgment? 
• Which decisions happen repeatedly? 
• Where could AI safely act on its own? 
• Where should a person always remain involved? 
• How will we know whether the redesigned loop is measurably better? 

Then redesign from there. 

One of our favorite questions for transformation work is: “If we designed this process today, knowing what AI can now do, would we design it the same way?”

Increasingly, the answer is going to be no. But that does not mean the answer is simply “automate it.” It means we have an opportunity to rethink the roles of people, AI, software, data, and automation together.

What is the bigger opportunity?

For years, enterprise transformation has largely involved implementing technology and teaching people to work within it. Much of our industry has been built around that model. 

That shift also changes enablement. As AI and digital adoption platforms become more embedded in the flow of work, helping people adapt becomes less about training them outside that flow and more about supporting better decisions as the work happens. 

AI gives us an opportunity to start reversing that relationship. Instead of designing work around the limitations of our applications, we can increasingly design technology around the outcome people are trying to achieve. AI can take on more of the research, analysis, monitoring, coordination, and routine execution. Traditional software can continue doing what deterministic software does very well. And people can spend more of their time on higher-value work:

• Judgment calls 
• Relationship development 
• Creative work 
• Previously unsolved problems 

That is why we think Atlassian’s announcement is worth our attention. It treats AI as a participant in a continuous, governed operating loop, with a person still leading the system.

Agents do the tasks. The governed loop turns those tasks into an outcome, with a person leading it. 

The companies that create the most value from AI will be the ones that get very good at deciding where people should lead, where AI should assist, where AI can act, and how all of it comes together to produce a better outcome. 

That, to us, is the next phase of enterprise transformation.

Let’s talk about your loops.

At Collective Insights, we help organizations rethink how work gets done as people, AI, SAP, Digital Adoption Platforms, data, and automation come together. We identify where AI can improve performance, where people need better in-the-flow support through tools like WalkMe, and how enterprise processes can keep improving.

If you are modernizing SAP or rethinking how work should happen across the enterprise, we would welcome the conversation.

Source: Atlassian, “We’re bringing governed agent loops to the AI-Native SDLC,” September 10, 2026. Atlassian currently describes Agent Loops, Standards, and AI Review as private early-access capabilities.