When Will Agentic AI Cross the Chasm?

In 2022, ChatGPT, leveraging Large Language Models, swept us all up in a whole new vision of what generative AI could be.  It has enormous potential to create ROI from amplifying high-value human-in-the-loop processes like wealth management, patient diagnosis, campaign development, and the like. 

By contrast, agentic AI, which is just now coming to the fore, has the goal of taking humans completely out of the loop.  To date, people have been focusing on how this will reduce labor costs, which is both a real and socially consequential outcome.  That said, it will be nowhere near as consequential as the step-function improvements in response times and error rates that agentic AI will enable.  Whenever you take a material amount of time and scrap out of any process, it changes the dynamics of the entire system and releases trapped value you did not even know was there.  Unlike generative AI, in other words, which can be expected to create billions of dollars of ROI, agentic AI has the potential to create trillions—once it has been deployed successfully!

And that brings us to the present day.  Agentic AI has yet to even cross the chasm, much less get inside the tornado.  So, when is that going to happen, and what is that going to take?  Here’s my read:

1. To get the ball rolling this year, agentic AI vendors need to target the most impactful value-trapping bottlenecks in the industries where they are causing the most pain.  These would include claims and billing dispute processing, credit authorizations, pricing approvals, medical authorizations, regulatory compliance reviews, and the like.  What they all have in common is that they suck up an enormous amount of human resources and time to deliver what are atomically low-value transactions.  Moreover, each holds an end-to-end value delivery process hostage while it works its way through its time-consuming steps.  At the end of the day, however, all these transactions are grounded in processing a linear series of algorithms, such that AI can execute them in seconds, whereas humans take minutes, hours, days, weeks, or even months.  Breaking these time bottlenecks is how the trapped value gets released.
2. To deploy at scale in the near term, this first wave of solutions must come with guardrails already built in.  That means they must be deployed atop enterprise applications already proven to be reliable—notably the ERP and CRM systems that are the backbone of today’s global commerce.  The only change will be that instead of humans invoking them, agentic AI algorithms will.  This stands in stark contrast to DIY efforts underway in many enterprises today, where data scientists working with the state-of-the-art software seek to build next-generation transaction flows.  The demos work great, but because they are not leveraging decades of real-world testing, they inevitably hit serious deployment roadblocks when they go to scale.
3. On the customer side, the disruptive nature of agentic AI applications warrants them being hosted in the Incubation Zone, not in the Productivity Zone.  That is, they call for a focus on agile development, restricted initial deployment, and iterative adjustments, all in preparation for a transition to the Performance Zone.  Because these applications are not treading any new ground, but rather accelerating transactions that are already well tested, they do not require a Transformation Zone approach to taking them to scale.  Instead, a thoughtful, well-monitored ramping up should do the trick.
4. Finally, let me give a call out to both vendors and customers to stop talking and start doing.  The processes we are talking about are known bottlenecks in core processes that gate the productivity of entire industries.  Everyone agrees that the customers of these industries are not being as well served as they could be.  It behooves us all to get off our duffs and, as the Nike folks like to say, Just do it!

That’s what I think.  What do you think?