I recently spoke with a leader in a very large services organization about their experience adopting AI. They told me they struggle most when, even before their latest AI project is operational, a new technology arrives and the team is pushed to pivot and try to take advantage of it. This kind of self-disruption is an all-too-common occurrence nowadays. It’s like building a house where the foundation keeps changing shape before you can get the roof on.
The battle to finish what you start is being lost to the fear of missing out.
All your estimates are wrong
In my last post, I suggested that quality data is the foundation for successful agentic initiatives. With this principle in mind, how do you address the ephemeral nature of agents? They come and go so quickly. Investments in development and process change hardly have time to take root before they are eclipsed. Tried-and-true project estimation and budgeting methodologies fail.
Years of education and experience running IT projects did not prepare you for this pace.
There is a line that I call the “ROI horizon.” It’s the point in the timeline of a technology lifecycle where it is possible to measure the outcome of the effort rather than just the cost. If your effort does not survive long enough, that is a line you may never reach.
Understand the AI agent lifecycle
Hyperscalers and model creators have given us powerful tools that allow us to iterate at warp speed. The pseudo-technical are now empowered to create highly capable components that contribute to our work and make us more efficient. I am constantly called aside by people to see the new Gem (custom AI agent) they created. It truly is amazing. Except there is a challenge.
That thing we once called "knowledge work” is not stagnant. It evolves constantly with economic conditions, org structures, new technology, and sometimes, yes, even the weather.
Our systems have been codifying business rules into applications that govern processes—that is, the work itself—since the dawn of IT. Applications assume the responsibility of recording what happens and moving things from one step to the next. But there has always been a flexible and reactive actor in the workflow: the human. Early on, if you had an accounts payable system and an order processing system, the very first integrations were in the form of keystrokes—until it became financially beneficial and technically feasible to make the machine do it alone.
This was a slow and expensive evolution.
Those humans had careers that spanned decades. Agents allow us to take over tasks that would take humans exponentially longer to finish—tasks requiring skills that would take them years to learn. But unlike humans, most of the agents that we create today are capable but not truly adaptive. We use them and are amazed, but in just a few weeks, something changes. Adaptation is necessary, and more often than not, we replace the agent we created six months ago with a completely new one.
Do you even bother to measure the return on investment for something that you have already replaced? How sure are you that it was worth it?
The AI agent lifecycle forces leaders to rethink all aspects of business optimization. Integrating two systems can now happen in hours instead of months. But you are also likely to replace that solution much more quickly. This not only disrupts how the work gets done, but perhaps more importantly, it overturns the calcified enterprise budgeting cycle.
The data layer is the long game
From an investment perspective, it makes sense to first ensure that the data foundation is solid. Next, you should focus on that data architecture’s support for the rapid evolution of the agents that feed on it. This is where Model Context Protocol (MCP) has come into play. MCP becomes the flexible gateway that the constantly mutating soup of agents uses to access the data that they need. The agents change, the access layer becomes more fluid, and the data provides stability as it is created and moved through its various states through an evolving business process.
Probabilistic interactions age agents faster
Not all agents age at the same rate, and understanding what accelerates their obsolescence can help you prioritize where to invest.
A given agent tends to live longer when it is dealing with a deterministic flow. Automating well-understood, stable fragments of mature workflows produces the types of agents that naturally become incorporated into core systems. The ability to handle probabilistic scenarios is the most interesting and transformational.
We constantly hear promises of the magical thinking agent that will replace your analytics staff. It is not likely to be effective in the near term. The rate of change in the conditions in which they exist shortens their viability. Be wary of solutions that overpromise on autonomous analytics and underdeliver on an AI-ready data architecture.
The bottom line
At Fullstory, we are investing in the concept of agentic workspaces to address a variety of key jobs to be done. It is an inevitable evolution, but rest assured, we are building that capability on the foundation of Fullcapture.
We began by launching Fullstory’s StoryAI to put a variety of AI capabilities into the hands of our customers. Our next leg in the agentic race is opening our data up so that it is consumable by agents across our customers’ ecosystems as they evolve.
Fullstory MCP and other developer-focused initiatives, like our Fullstory Skills Repository, are both enabling the evolution of agents that you and our partners will create with Fullstory data.
These efforts are the critical pillars you need for analytics that will live long enough to see their value realized. Agents—especially those created by individuals—are ephemeral. Built in hours today. Quickly replaced tomorrow.
Embrace the pace of change. Invest in a stable, data-driven foundation. Measure the value of your efforts from the outcomes. Accept that self-disruption is part of the formula now, but endeavor to push it "above the data line" to extend the lifespan of your investments past the agentic ROI horizon.
