Most enterprises have spent the last two years standing up AI pilots, experimenting with copilots, and asking their architects a version of the same uncomfortable question: what do we do about all the old stuff?
On a recent podcast with Mike Vizard for the Techstrong AI Leadership Insights series, Jeremiah Stone, CTO of SnapLogic, asserts that the question is usually asked too late and framed incorrectly.
“Legacy” and “Heritage” are not the same problem
Before you can build an AI integration strategy, Stone argues, you need to be honest about what you’re actually working with. And that starts with a distinction most organizations are quietly fumbling: not all old systems are the same problem.
A 25-year-old ERP that runs your entire order management process is fundamentally different from a custom-built reporting tool nobody knows how to maintain anymore. One is a foundation carrying decades of business logic and domain expertise. One is technical debt. Treating them the same way (replacing everything, or protecting everything) is one of the most reliable ways to stall a transformation program before it starts.
Stone’s framing is essentially a triage argument. Before you build an AI integration strategy, you need an honest map of what you have, what it’s worth, and what role it should play going forward. He’s seen this movie before. He entered the industry during the late 1990s shift from mainframes to personal computers, which was the last time organizations had to figure out how to distribute powerful new capabilities across infrastructure built for a completely different era.
“The companies that navigated it well weren’t the ones that replaced everything,” he says. “They were the ones that bridged intelligently.”
Jeremiah Stone, CTO, SnapLogic
That’s still the right instinct now.
Where AI is actually earning its keep
There’s a useful filter Stone applies when he’s evaluating where AI investment is actually worth making, and it starts with an honest look at the process you’re trying to improve.
Not all processes are the same. Some are deterministic: clean, rule-based, well-understood. Run the same inputs through them a thousand times, and you get the same output. Others are inherently probabilistic: messy, judgment-intensive, and dependent on experienced individuals making decisions under uncertainty.
Stone’s observation is that AI tends to deliver its most meaningful returns in the second category, not the first. Layering AI onto a process that already works predictably rarely moves the needle much. Applying it to the kind of work that was always a little chaotic (where humans were already improvising) is where cycle times actually compress and quality improves.
Those efficiency gains are real. But they’re also what every competitor is pursuing simultaneously, which means the advantage compresses almost as fast as it appears. The more important question is what comes after the efficiency wave.
Take, for instance, Uber. Before Uber, if you wanted a car, you called a number, hoped someone answered, gave an address, and waited with no reliable sense of when anything would arrive. Uber didn’t just make that process faster. It built something that hadn’t existed before: real-time matching of supply and demand at city scale, with reliable ETAs, dynamic pricing, and driver accountability baked in. That’s not an efficiency gain. That’s a new category of service.
“That moment is still to come for language model-based systems. Where it’s really happening right now is in the software development space. But more broadly in the economy, I think it’s still early days.” – Jeremiah Stone, CTO, SnapLogic
It’s a more honest framing than most AI commentary offers right now. The operational efficiency wins are worth pursuing, but they’re not a sustainable advantage. The organizations that pull ahead will be the ones that use AI to build services that genuinely couldn’t have existed before. And that requires a different kind of ambition than the current wave of productivity optimization.
Register for AgentFest Virtual Summit 2026 and learn about orchestrating agents, data and MCP at scale.
The leadership demand is going up, not down
There’s a version of the AI narrative that goes: AI handles the routine, humans focus on the creative and strategic, everyone wins.
Yes, AI is accelerating the empowerment of the skilled individual. Administrative overhead is compressing. The leverage available to a strong engineer, analyst, or operator is genuinely expanding. But Stone pushes back on the idea that this reduces the demand for leadership. He thinks it increases it.
“I don’t think there’s ever been a greater need for good leadership and guidance when you have such a fast, dynamic market. The burden and pressure on consistent direction, clarity of goals, and execution is more profoundly felt than ever.”
Jeremiah Stone, CTO, SnapLogic
When the pace of change accelerates and individual leverage increases, the cost of unclear direction multiplies. A confused team moving slowly is recoverable. A confused team moving fast, with AI amplifying their velocity in the wrong direction, is a much harder problem.
The organizations navigating this well aren’t the ones cutting headcount as AI automates tasks. They’re the ones redeploying experienced people toward higher-leverage work, while building the kind of clear strategic context that lets that leverage land.
The constraint on AI adoption is strategy
The thing Stone keeps returning to, in different forms, is that the constraint on enterprise AI adoption isn’t the technology. The models are capable. The tools exist. The integrations are buildable.
The constraint is the quality of the decisions being made around the technology, whether:
- You can distinguish a heritage system from a legacy one
- You’re targeting AI at the processes where it actually moves things
- You’re building toward new value rather than just optimizing what you already have
None of that is an AI problem. It’s a strategy and organizational problem that happens to have AI sitting at the center of it right now. The companies that get those calls right won’t necessarily have deployed more AI than their competitors. But they’ll have a clearer sense of why they deployed what they did. And that turns out to matter quite a lot.
Jeremiah Stone is the CTO of SnapLogic. This post is based on his conversation with Mike Vizard for the Techstrong AI Leadership Insights podcast, “Why Legacy Systems Are the Hidden Constraint on AI Agent Adoption”, published February 2026.
