AI agents can autonomously complete complex tasks without a predefined workflow. As such, they promise to transform everything from software development to the way entire organizations operate. And with the very recent release of ChatGPT agent, which can proactively choose “from a toolbox of agentic skills to complete tasks for you using its own computer,” per OpenAI, agentic AI is rapidly approaching general availability.
Still, when it comes to realizing the promise of agentic AI, we are very much in the early days. For example, a recent CapGemini report on the rise of agentic AI found that only 2% of organizations had deployed AI agents at scale (with another 12% saying they had done so “partially”). The reasons for the slow pace of adoption are several, but one need look no further than some of the inherent challenges the technology faces today.
Engineer and researcher, Utkarsh Kanwat, describes some of these challenges in an essay he wrote explaining why he is “betting against AI agents in 2025.” The challenges include: compounding errors across truly complex, multi-step tasks; context windows rapidly increasing token costs; and the lack of tools and systems that agents can easily use.
None of these problems are in and of themselves insoluble. One can, as Kanwat describes, break tasks down into more discrete units that agents can easily and reliably handle. What’s more, protocols have already appeared to facilitate agent-to-agent interaction. And one can conceive of orchestration patterns that reduce the context burden for individual agents.
All these issues aside, however, there is another, larger problem organizations must confront: They are simply not yet ready to support the widespread use of agents.
A recent article on Accenture’s development and deployment of its Advanced Technology Agent highlights the homegrown issues organizations looking to adopt agentic AI face.
First, as Steven Courtney, a managing director at Accenture said, “One of the biggest challenges we found in terms of making this happen on the ground was data and data accuracy.”
Since Accenture aimed to use their agent to manage IT operations, they had to ensure that the data from various systems was accurate and secure. This was not always the case.
Second, there was the problem of process design. Rajandra Prasad, CIO and asset engineering officer at Accenture said, “You need to get your business processes reinvented before you infuse technology. If you have a very inefficient business process, adding very high-powered agentic tech to that inefficient process just makes your inefficiency run more efficiently.”
In other words, as powerful as they may be (or will someday become), AI agents are not magic bullets capable of solving existing problems related to organizational visibility and real-existing operations. As with any cutting-edge technology, one cannot assume that the organization is ready for it. Since AI agents appear to be the future of software (and much, much more), how does an organization become future ready?
Getting the current state right
The short answer to that question is this: The organization must transform itself.
Since the impact of agentic AI, and AI more generally, will come in waves, the reality is that organizations must become capable of transforming again and again, both in individual functions and across the entire organization. In other words, transformation must become a capability, nurtured and strengthened through repeated use.
These transformations will take place across three fundamental dimensions: the processes that run your business, the technology those processes run on, and the people using this technology to execute those processes. The foundation of this capability thus becomes consistent, accurate, and reliable insight into the as-is state of your processes, applications, and the digital adoption of technology by your employees.
Several practices and tools designed to provide the relevant visibility already exist.
Business process management relies on process mining tools to understand how processes actual operate in the enterprise. These tools can uncover common workarounds and variations in documented processes, as well as help you benchmark processes against industry standards. The insights thus provided are key to process optimization, allowing you to identify and act on areas for process improvement. These insights also serve as the necessary starting point for inserting agentic AI into your business processes.
Your processes depend on underlying technologies, the applications that facilitate supply chain management, human resources management, manufacturing, and so on. It is these applications that AI agents will need to access and use if they are to take over specific tasks and workflows. A reliable inventory of these applications along with an understanding of how they interact – what depends on what? where is data produced? where consumed? – is a prerequisite for agentic AI adoption.
Enterprise architecture management and related toolsets handle the creation and maintenance of this inventory. In fact, a survey we conducted revealed that creating such an inventory is the top priority for most enterprise architects.
The enterprise architecture practice will also oversee the continuous rationalization and rightsizing of the application landscape. Such stewardship of the IT estate is critical as the technological needs of the organization continuously evolve with the introduction of agents and a pivot towards a toolchain built for and around agentic AI.
Finally, you must account for the current state of the people dimension. Traditionally, companies have invested in enablement and training to facilitate and manage technology adoption. Unfortunately, this has not always been adequate, primarily because a training-first approach does not afford the necessary insight into actual usage, including common bottlenecks. With the right digital adoption platform, one can not only track usage, but, more importantly, identify friction points in technology adoption that either degrade productivity or hinder adoption altogether.
Digital adoption platforms serve two purposes in creating the future-ready organization. First, they accelerate adoption and can even provide an initial layer of automation for repetitive and predictable tasks. Second, they can facilitate interactions between people and the agents they work with.
“But I thought agents were supposed to be autonomous,” you might say. And, yes, autonomy is one central promise of agentic AI. But, the fact of the matter is, humans will have to stay in the loop. Trust and risk management is one reason for this. The other is that humans are very good at understanding context, especially when informal or intuited. As the tasks agents take on become more complex and context-dependent, people will increasingly take on the role of providing guidance, correction, and formal approval of agentic actions.
Managing the future
Preparing your organization for the future doesn’t stop with getting your house in order, as just described. You also need to be able to manage the future state, and that itself calls for preparation.
As you begin to adopt agents, they will increasingly become your current state. Accordingly, you will want to observe and manage this adoption. That will involve being able to answer some questions about your agentic landscape:
- What agents do we have currently available in the organization?
- Much like your application inventory or an inventory of services in your product organization, you will want an inventory of your agents, both those purchased or rented from third parties and those you have built yourself.
- Like those other inventories, you will want to add context including the business capabilities the agent supports, the process(es) it serves, the applications it accesses, and the data it produces/consumes.
- Finally, you will want to rationalize this estate as well, ensuring that you are not duplicating efforts and identifying where you have gaps in supporting business capabilities or where you have duplications.
- How can we identify the best opportunities for agentic deployment?
- Finding the right agentic use cases is one hurdle companies encounter today, and it in part explains the limited deployment of AI agents in the market.
- However, even as the use of AI agents matures and best practices emerge, the pursuit of new use cases will not end. Indeed, as the technology becomes capable at working reliably at higher levels of abstraction, there is no hard limit to what agents might eventually undertake.
- As it turns out, agents can help in this instance, especially when applied to process mining and analysis. That is, agents can already identify areas for potential improvement in business process and recommend alternatives.
- For maximum effectiveness here, agents will need data and a contextual view. Digital adoption platforms can provide both by uncovering areas where AI agents can effectively support and by helping manage the interactions with agents within workflows.
- How effective are my AI agents?
- AI agent adoption is not an end in itself. Companies expect their deployment to increase efficiency and productivity. In cases where agents are deployed in customer-facing roles, they can even drive revenue growth. Whatever the goal, just as with any other technology, you will want to track and measure whether an agent or a network of agents is delivering the intended value.
- Collecting the necessary information to evaluate agent performance relies on the various practices mentioned above. The AI agent inventory tells you what agents to track. Your process management practice can measure the performance of agents in context. And your digital adoption platform can show the impact of agents on the performance of individual users.
- What impact do my agents have on the skills of my workforce?
- Although there have been a lot of high-profile claims regarding AI agents replacing people in the workforce, the more likely scenario is that AI agents will change the portfolio of skills needed by the organization.
- These changes in sought-after skills will have an impact on the recruitment front by transforming the types of roles you need to hire for and the skills those new hires will need. To pick just one example, consider the following: if AI agents can automate many elements of the BDR function, then your go-to-market organization may end up needing more sales engineers and account executives.
- At the same time, as AI agents become more integrated into workflows, your team will need training on best ways to interact with agents and even how to build their own.
Tomorrow starts today
By now it should become clear that, no matter where you are on the agentic AI learning and maturity curve, there are things that you can do today to prepare for what’s to come.
For starters, you can focus on mastering the “as-is.” Most companies today struggle to catalogue their IT estate, map existing business processes, or track usage of different software applications. If these are immature or informal practices within your organization, this is where you start. Frankly, the day you want to launch your first AI agent POC, you’re going to wish you had already done this.
While you are doing all that, you can put in place rules for agent governance. Who can build agents? How do agents get approved? What types of data can agents work with? What types of jobs or data are off limits? Your organization will already have rules for architecture governance, data governance, and the like. Similarly, Agents will need ground rules, guidance, and governing bodies. These should be developed and implemented before any agent is released into the wild.
Finally, you should focus on developing AI literacy within your organization. When employees know what agents are and what they can do, they will themselves uncover new use cases and innovative ways to employ them.
There is a lot of hype around AI agents. And some of the boldest claims – “AI agents can reorganize your supplier landscape in real time in reaction to tariffs” – if not farfetched are certainly still far off.
Nevertheless, the genie is out of the bottle. AI agents are here and will only get better. Your time to prepare, so your organization can get better along with them, is now.
