Vibhor Kumar is a technology executive and author focused on enterprise data and AI platforms, modern architecture, and technology strategy.
Technology leaders are constantly asked to make long-term decisions with short-term information. Choose a cloud platform. Select a database. Standardize an architecture. Adopt an AI platform. Each of these choices looks defensible the day it’s made. The real test comes years later, when the business has changed shape, and the assumptions behind the original decision stop holding.
That’s the uncomfortable truth I keep coming back to after two decades working with banks, insurers and telecom operators on their data platforms: the most expensive technology decisions aren’t the ones made today. They’re the ones organizations are still living with a decade from now.
AI has only sharpened this. Models that look dominant today can lose their advantage surprisingly quickly. Agent architectures are still being invented, not just adopted. Trying to predict the technology stack of 2036 is the wrong goal. The better question is whether the decisions being made today preserve the ability to change tomorrow.
A Story I’ve Seen Play Out More Than Once
Consider a large bank that made a major platform decision more than a decade ago. At the time, it was a well-reasoned choice: proven, highly available, well understood, capable of carrying mission-critical workloads.
Over the following years, more applications were built on top of it. More operational processes wrapped around it. What started as a technology choice slowly hardened into an architectural dependency, not through any single bad decision, but through a decade of reasonable ones. Then the bank’s priorities shifted, as they always do. Cloud became part of the strategy. Real-time analytics needed to sit closer to the data. AI introduced new demands for enterprise context. Regulatory expectations around resilience and data sovereignty kept moving.
The original platform hadn’t failed. It had done exactly what it was chosen to do. The problem was that changing it had become enormously difficult. Moving away wasn’t a database migration anymore; it meant untangling years of applications, integrations and assumptions built around one decision made a decade earlier.
This is what long-term technology cost actually looks like: It never appears on the original procurement document. It accumulates in the dependencies that build up around a decision long after the decision itself is forgotten. The lesson isn’t that the bank chose wrong. It’s that the more useful question was never “Is this the right platform today?” but “Does this preserve enough optionality for when our requirements inevitably change?”
Three Questions To Ask Before A Long-Term Technology Commitment
Consider these questions before making any long-term decisions.
1. Does this architecture preserve optionality?
A lot of technology strategy gets built around forecasts—which cloud will dominate, which model will win. Predictions expire; optionality compounds. Rather than standardizing deeply around whichever model performs best on a workload right now, the better question is whether a better one can be adopted later without redesigning everything around it. The same logic applies to databases, clouds and analytics platforms.
2. Are you treating data as the durable asset?
Applications come and go. Infrastructure gets refreshed. Models turn over fastest of all. But enterprise data—customer history, transaction records, institutional context—routinely outlives all three. As access to capable models becomes more widespread, enterprise differentiation will increasingly depend on the quality, context and governance of an organization’s own data. Build applications and AI systems around the lasting value of that data, not the other way around.
3. How reversible is this decision?
Good architecture reviews rarely ask what happens the day an organization decides to stop using a platform. Before committing, it’s worth asking plainly: How hard would it be to move the data out, and how much business logic now lives inside the platform rather than in code the organization controls?
That second question matters more than it first appears. Data portability is not the same as architectural portability. A platform can make it easy to export the data while still leaving years of business logic, integrations, operational practices and institutional skills tied to that platform, which is closer to what happened at the bank than any single point of technical lock-in. Not every decision needs to be easily reversible; sometimes deep commitment is worth the tradeoff. But that should be a conscious choice, not something nobody notices until it’s too late.
The Takeaway
None of this argues against commitment. No organization can innovate while keeping every option open forever, and decisions have to be made. But leaders should know going in which of those decisions are easy to reverse and which may shape the organization for the next 10 years, because that distinction matters more with AI than with almost any technology cycle before it.
The next time a major technology investment lands on your desk, ask the usual questions: Does it solve the problem, does it perform, is it secure, what will it cost? Then add one more: If the world looks very different in five or 10 years, how hard will this decision be to unwind?
Nobody can predict exactly which technologies define the next decade. The job of technology leadership was never to predict the future perfectly—it’s to build an organization that can adapt when those predictions turn out to be wrong.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?