This is part two of our 6 part series on AI's adulting phase. If you missed the last one, read it here

As we started exploring last time, the latest shift in AI is agentic AI, where different AI systems pass tasks back and forth between each other until they find a solution. Picture a software development team where AI agents might pick up a ticket from a bug report to fix the issue. One agent serves as the debugger, another as the code writer, one as the refiner, and another as the tester. They pass work between themselves with the promise of replacing whole teams of people.

But there's something more immediately useful happening right now in business—what I call "outsourced action" rather than fully agentic AI. Instead of a team of AI agents working completely autonomously, we're seeing workflows where AI systems handle specific steps before passing control back to humans.

The Agent Team in Action

Let me paint a picture of how this actually works. Imagine you have a folder in SharePoint. Every time you add a document, the system automatically kicks into gear. AI picks up the file, reads it, summarises the content, and condenses it into a draft LinkedIn post or blog article. That draft then gets handed to another agent that creates a branded image and assembles the complete content package.

These aren't fully autonomous systems—they're more like intelligent workflows that use different AI capabilities to move work to specific checkpoints where human oversight becomes necessary again. It's practical, it's happening now, and it's genuinely useful.

This represents a fundamental shift to what Gartner calls the "paramount strategic technology trend" of 2025: agentic AI systems that can understand goals, plan actions, and execute them with minimal human intervention. Imagine having a "virtual workforce" of AI agents that can assist, offload, and augment the work performed by humans. Quite the promise.

Current Reality vs Future Promise

Let's be honest about where we actually are versus where the marketing suggests we're heading. Current AI agents exhibit what researchers call "jagged intelligence"—they can be remarkably capable in some areas while completely failing in others. This inconsistency creates significant challenges for organisations trying to deploy autonomous systems at scale.

Most successful implementations today involve "structured human oversight." AI handles the heavy lifting—data processing, initial analysis, content generation—but humans remain in the loop for critical decisions and quality control.

Consider a practical example: an insurance company might use AI agents to process initial claims, gather relevant documentation, and prepare preliminary assessments. But the final approval still requires human judgment, particularly for complex or high-value claims. This hybrid approach leverages AI's speed and consistency while preserving human expertise for nuanced decision-making.

Building Practical Workflows

The most effective implementations I've seen start simple and gradually build complexity. Rather than attempting to automate entire processes immediately, successful organisations identify specific steps within existing workflows that benefit from AI intervention.

Take content creation as an example. Instead of asking AI to produce finished articles, you might design a workflow where AI handles research and initial drafting, humans provide strategic direction and editing, and AI manages formatting and distribution. Each step has clear handoff points and quality checkpoints.

The key insight is that effective human-AI collaboration requires intentional design. It's not enough to add AI tools to existing processes. We need to rethink how work flows between humans and artificial intelligence to maximise the strengths of both.

The Infrastructure Challenge

One significant barrier to the widespread adoption of agentic AI is the need to address infrastructure and legacy systems. Many organisations lack the technical foundation necessary to support sophisticated AI workflows. This includes everything from data architecture and API connectivity to security frameworks and governance protocols.

The most successful implementations often start with organisations that already have robust digital infrastructure and clear data governance practices. These companies can more easily integrate AI agents into existing systems without compromising security or compliance requirements.

Navigating the Risks

The push towards agentic capabilities brings genuine risks that organisations must address proactively. When AI systems operate with greater autonomy, the potential for errors or unintended consequences increases significantly. This is particularly concerning when these systems interact with external stakeholders or make decisions that affect business outcomes.

Robust guardrails become essential. This involves implementing monitoring systems that can detect when AI agents operate outside expected parameters, establishing clear escalation procedures for edge cases, and maintaining human oversight for high-stakes decisions.

The challenge is balancing autonomy with control. Too much oversight negates the efficiency benefits of agentic AI, while too little oversight creates unacceptable risks. Finding this balance requires careful consideration of each use case and its potential impact.

Your Next Steps

Here's a practical exercise to help you explore agentic possibilities in your own work. Start by mapping a routine process that involves multiple steps and different types of work. How you handle customer inquiries, process expense reports, or prepare weekly reports.

Identify which steps involve data gathering, analysis, or formatting—tasks that AI handles well. Then identify which steps require judgment, creativity, or relationship management—areas where human input remains essential.

Design a simple workflow that chains AI actions for the routine steps while preserving human control at critical decision points. Use tools like Microsoft Power Automate, Zapier, or even simple email rules to create these connections. Or simply copy/paste between the different apps while you test which parts of the workflow work well.

Test your workflow with low-stakes examples first. Pay attention to where the handoffs work smoothly and where they create friction. Notice what types of errors or edge cases emerge, and build appropriate safeguards.

Looking Ahead

The trajectory towards more sophisticated agentic AI seems inevitable, but the timeline and specific implementations remain uncertain. What's clear is that organisations that begin experimenting with structured AI workflows now will be better positioned to leverage more advanced capabilities as they emerge.

The shift from AI as a tool to AI as a collaborator represents one of the most significant changes in how we work since the introduction of personal computers. The organisations that navigate this transition thoughtfully—balancing efficiency gains with human agency—will likely find themselves with significant competitive advantages.

But this transition requires more than just technical implementation. It demands a fundamental rethinking of how we design work, measure value, and maintain human dignity in increasingly automated environments.

Looking even further ahead, I would like to leave you with this. In April of this year, a group of AI researchers, forecasters, and future scientists published a report aimed at forecasting scenarios in AI development for the next decade. They called it AI 2027, as the biggest shift is likely to happen in the next couple of years. Perhaps not surprisingly, the first steps they predicted have already played out. All major AI players have launched agentic AI. The report explores what might happen if we let the market alone drive the development of AI, how AI is ultimately a key question for national security and what that might mean in the current political climate. Trigger warning: both scenarios they paint aren't very positive. And critics of the report say it still might be too optimistic. Food for thought. If you're more of a visual learner, this YouTube video does an amazing job of explaining it in visuals.

Next time, we'll explore what happens when this automation mindset meets the reality of human work and why the customer service evolution serves as both a cautionary tale and a roadmap for better approaches.

This is the second post in a six-part series exploring AI's transformation from hype to practical implementation. Next time: "Redefining Work: The Dumbing Down Dilemma."

PS: I usually like more positive header images, but I found this to be quite telling. It was the first image Google Gemini created when I asked it to "help me create an image for the theme "When AI Stops Waiting".

References

• Deloitte. (2024). State of Generative AI in the Enterprise 2024. https://www2.deloitte.com/us/en/pages/consulting/articles/state-of-generative-ai-in-enterprise.html
• MIT Sloan Management Review. (2025). Five Trends in AI and Data Science for 2025. https://sloanreview.mit.edu/article/five-trends-in-ai-and-data-science-for-2025/
• Gartner. (2025). Explore Gartner's Top 10 Strategic Technology Trends for 2025. https://www.gartner.com/en/articles/top-technology-trends-2025
• The Sirocco Group. (2025). The rise of Agentic AI: From generation to action in 2025. https://www.siroccogroup.com/the-rise-of-agentic-ai-from-generation-to-action-in-2025/
• Salesforce. (2025). Salesforce AI Research Details Agentic Advancements. https://www.salesforce.com/news/stories/ai-research-agentic-advancements/
• Daniel Kokotajlo, Thomas Larsen, Eli Lifland, & Romeo Dean. (2025, April 3). AI 2027. https://ai-2027.com