If, like me, you work in the digital consulting world, you likely spend a lot of time these days thinking about how agentic AI can benefit other businesses (specifically, those of your clients).
But the ultimate proof of agentic AI’s value, perhaps, lies not in showing how it can help other companies, but in implementing it for your own.
That’s precisely what my own organization, OSF Digital, is now doing as we roll out an agentic “operating system” for internal use. Our efforts are, I think, a prime example of how agentic AI technology can deliver benefits in the real world, as well as the unexpected hiccups organizations may encounter as they deploy AI agents into production.
See also: AI’s Flash Problem Hints at A Deeper Struggle in IT
The need for an AI agent “operating system”
We conceived our agentic operating system to help address one of the core challenges surrounding AI agents in businesses: How to incorporate them practically and pragmatically into business workflows.
After all, it’s easy to recognize the value of AI agents in isolation. They can run commands, make API calls, write code, generate reports, and so on.
That’s great – but it doesn’t translate to business value on its own. What organizations really need is a realistic, scalable means of leveraging agents to address everyday business needs, while also ensuring that they can sufficiently control the agents to prevent risky behavior, protect sensitive data, and so on.
Our agentic operating system (which we’re calling Kepler in honor of the Renaissance astronomer who helped pioneer modern understandings of planetary movements) solves this problem by connecting AI agents, models, skills, and orchestration tools into streamlined workflows.
See also: Exploring the Sovereign AI Aspects of Inference Servers
Accelerating digital transformation with agentic AI
To be more specific about how Kepler works and which problems it solves for my organization, let’s discuss the particular use cases it supports.
At my consulting company, we help clients undertake digital transformation projects. Historically, planning and implementing the transformations required a great deal of manual effort. We set up workshops with clients to assess their needs, then summarized the interviews into manual reports, then created user stories around them, then built the technology products necessary to implement those stories, then validated it all. The process could take months.
We knew that AI agents offered opportunities for accelerating these workflows. But we needed agents skilled in multiple domains (like assessing customer business needs, assessing technical requirements, and generating code, all of which are pretty distinct processes). We also needed ways of unifying the agents into cohesive workflows, while also ensuring that our human consultants remained the ultimate drivers of the work produced by the agents.
Kepler helps solve these needs by making a suite of agents and skills available to our consultants. The agents can automatically interpret client needs based on interviews and documents supplied by the clients. They can then draft user stories, plan transformations, and implement changes. As a result, transformation workflows that once took days now take just hours.
Also critical is Kepler’s support for “memory,” meaning its ability to record and remember data as it appears over the course of the transformation process. In this way, Kepler serves as a dependable, centralized point that all team members can use to understand the history and context of our customers. This is a big deal during complex projects that involve dozens of stakeholders because it means they no longer have to rely on human memory and manual knowledge sharing to preserve and transfer information.
We like to believe our agentic operating system is so powerful in part because it’s a platform designed by agentic technologists, for agentic technologists. In other words, our own consultants, whose jobs center around helping clients transform their systems and processes with help from AI, built Kepler by leveraging their agentic experiences to address our business’s own internal needs.
AI agents as human accelerators, not replacements
Probably because Kepler was built for AI experts, by AI experts, a key aspect of the system is that it doesn’t try to make humans irrelevant during transformation projects. On the contrary, we designed it so that human input and guidance are core elements of agentic workflows.
In effect, Kepler only generates drafts of customer requirement assessments, user stories, and technology products. Our consultants retain the ability to modify (or, in extreme cases, veto) what agents propose. They can also tell agents to reformulate results based on new criteria or different areas of emphasis.
This is all critical because it’s hard to imagine a future where even the best AI models and agents are capable of managing large-scale projects on their own. They’re complements to human expertise, not a replacement.
Conclusion: Putting agentic AI theory into practice
To summarize, a real-world approach to unlocking agentic AI’s value must center on identifying actual business problems, then building unified agentic platforms that can address them – while ensuring that humans remain in control of the overall process. That’s how organizations can bridge the gap separating the theoretical benefits of agentic AI from actual value.
