A recent Fortune article highlighted a growing problem inside organizations aggressively adopting AI: employees are producing what Shopify CEO Tobias Lütke calls “slop grenades” – AI-generated emails, code, and other outputs that look complete but ultimately create more work for the people receiving them.
Industry analysts have given this phenomenon a name: “workslop.” The output may look polished, but it lacks the context, accuracy, or substance required to actually advance the work. Instead, someone downstream has to interpret it, correct it, or start over.
It would be easy to look at this as evidence that organizations have moved too quickly on AI. However, we think the more important lesson is how AI capabilities are built and operationalized within organizations.
That distinction is becoming increasingly important as organizations move from experimenting with generative AI to deploying agentic systems capable of performing increasingly complex work at scale – where the risks of generating large volumes of unusable outputs can be both costly and crippling.
Access to AI is not the same as an AI-enabled operating model
Many organizations began their AI journey by putting copilots and large language models in employees’ hands and asking them to find ways to become more productive. That was a reasonable place to start – but access to AI is not the same as an AI-enabled operating model. Our 2026 Acxiom–Forbes Insights survey points to this operationalization gap: 53% of business leaders say their current identity framework supports AI-driven and agentic marketing only in limited use cases, while 36% say it cannot support these efforts effectively today.
When organizations simply add AI to existing tasks, they may improve efficiency at the margins without addressing the underlying operational challenges. In some cases, they may even create new ones: employees complete the same work in a different way, while downstream teams must review, interpret, correct, or recreate the output.
The real opportunity is not simply to give people more AI tools. It is to redesign how work gets done based on what AI can do well. That requires organizations to rethink processes, roles, decision rights, data, technology, and governance.
The mandate is to design AI capabilities with the right data, context, guardrails, workflows, integrations, and human oversight – while also rethinking how teams operate to take advantage of AI’s strengths. Done well, this approach can help organizations move beyond incremental productivity gains and create more connected, scalable, and measurable business impact.
AI needs organizational context to automate organizational work
Agents can reason, plan, call tools, interact with systems, and execute increasingly sophisticated workflows. But without knowing how your organization needs to operate, the decisions agents produce may not align with your operational objectives.
For instance, we’re developing an agentic capability intended to help enterprise marketing organizations streamline content compliance processes. We could simply feed the agent documents outlining federal and state regulatory statutes, but organizations may interpret those statutes differently or have overlapping internal policies or disclosures that should be considered.
As such, the solution is to build capabilities into the agent that allow organizations to codify their internal requirements, specify rule interpretations, and document their feedback over time to ensure outputs are accurate, relevant, and performant.On another program, we were designing an agentic capability to create dynamic marketing workflows in Adobe Workfront, optimized for each campaign.
For a system like this to produce useful outputs, organizations need to capture and codify the institutional knowledge that humans intrinsically rely on: policies, roles and responsibilities, tool requirements, terminology, decision criteria, and other knowledge accumulated across the organization.
That information is then used as context for the AI system – structured, retrieved, interpreted, and applied at appropriate points in the workflow. Without it, the agent could still produce an answer, but it wouldn't necessarily produce the answer the organization needs – in other words, “workslop.”
This “context engineering” becomes one of the most critical enablers for successful operationalization of AI capabilities in enterprise environments – and is one of the first things we consider when designing new AI tools.
If these capabilities existed when we designed this process, would we have designed the process this way at all?
I saw this first-hand in a program I led with an energy company last year. The initial opportunity involved using generative AI to accelerate creative production. But simply generating more creative, faster, would not have solved the underlying business problem.
The organization needed to deliver more relevant customer experiences across a growing number of audiences, channels, and use cases. Producing more traditional, flat creative assets would only scale the constraints of the existing model.
So instead of simply inserting AI into the existing production process, we designed a new way of working enabled by the new AI capabilities. This involved moving them from producing flat, finished assets toward a modular, atomic content model that created reusable components that could be dynamically assembled and adapted for different audiences, channels, and experiences.
AI, along with other tech solutions, became the enabler of a broader system for creating, managing, assembling, and activating content.
AI systems need to be integrated, not just deployed
Producing an accurate AI output is only part of the equation. To create value at scale, that output also needs to move through the organization in a form that other people, systems, and processes can use.
In enterprise environments, that often means agents cannot simply return a block of text. Outputs may need to be structured against specific ontologies or schemas, enriched with metadata, routed into downstream platforms, used to trigger other workflows, or passed to another agent or system for the next step in a process.
This is where both context engineering and operating model design become especially important. If an agent misunderstands an organizational requirement, produces information in the wrong format, or routes an output incorrectly, the problem is no longer confined to an individual user reviewing a bad response. That error can propagate through an automated workflow, potentially creating “workslop” at scale.
The same is true of human oversight. Human-in-the-loop controls should not simply add another manual review step to every AI-enabled process. Organizations need to deliberately determine which outputs can move autonomously, which require approval, which should be reviewed by exception, and which conditions should trigger escalation to a human.
In a current engagement, for example, human reviewers are not only validating whether agentic outputs meet complex organizational requirements. Their decisions and feedback are captured as inputs into a continuous improvement loop, helping refine the context, rules, interpretations, and decision logic the capability uses over time.
The objective is therefore not simply to connect an AI capability to enterprise systems. It is to design an end-to-end operating model in which trusted outputs are structured correctly, routed appropriately, reviewed intelligently, and continuously improved as they move through the organization.
That is ultimately what allows organizations to move from isolated AI assistance to agentic capabilities that can reliably perform meaningful work at scale.
The workslop conversation is useful because it exposes a broader issue with the first phase of enterprise AI adoption.
Organizations have spent considerable time asking:
Where can we use AI?
The next phase requires harder questions:
- How should work be redesigned now that these capabilities exist?
- What organizational knowledge must be made available to AI systems for them to perform useful work?
- What tools and enterprise systems should agents be able to interact with?
- Where should humans remain in the workflow, and what governance, measurement, and continuous improvement mechanisms are required to generate scaled impact for the business?
There is no single blueprint for effectively operationalizing AI capabilities, but these questions are increasingly common across the organizations we work with. If you're working through similar considerations, Acxiom Adobe Consulting Services can help you assess the data, workflows, governance, and operating-model changes required to turn AI capabilities into measurable business value.
Subscribe to our blog
From the Blog
