The hidden architecture of disruption: What today’s founders can learn from 50 years of data integration

The startup world thrives on speed, novelty, and forward momentum. But the smartest founders — especially those building in AI and data infrastructure — understand that true innovation doesn’t start with a blank page. It begins with understanding what’s already been built, what refuses to break, and what’s quietly being eroded beneath the surface.

That was the thesis behind a wide-ranging conversation between Lonne Jaffe, managing director at Insight Partners and former CEO of Syncsort (now known as Precisely), and Tristan Handy, founder and CEO of dbt Labs, on Handy’s Analytics Engineering Podcast.

Jaffe, who began his career at IBM and now invests in deep tech companies, has spent years analyzing why certain technologies persist and others disappear — insights that are increasingly relevant in a genAI-powered world.

Here are five strategic takeaways for founders building the next wave of enterprise software.

Listen to their full conversation here.

1. Performance primitives never go out of style

In 1968, a company called Syncsort (which Jaffe later led) created a business around the technical operation of sorting, along with related primitives like merging, joins, and aggregations. At the time, sorting was more than just a database function — it was a foundation of mainframe computing workflows across industries like telecom and banking.

Fast forward 50 years, and sorting still anchors the logic of many data systems, from databases to Apache Spark.

“When Apache Spark came onto the scene, one of the first things the team did to show of its power and speed was to have it compete on the 2014 Gray Sort Benchmark,” Jaffe said. “Performance in sorting and related things like merging and joins and aggregation is still very, very important.”

The takeaway for founders? Building something new doesn’t mean ignoring what’s old. It means recognizing which technical primitives continue to drive performance and cost at scale, and optimizing where it still matters.

2. Switching costs are powerful — until they’re not

Jaffe’s career spans companies that both benefited from and attacked entrenched systems. That gives him a clear-eyed view on switching costs: They can be the strongest form of defensibility in enterprise software, but also the most brittle.

“You might have a situation where you ask a thousand customers, ‘Do you love this product?’ and 0% say yes,” he observed. “All of them would love to move off… but they don’t.”

Historically, moving from legacy platforms required rewriting massive codebases with minimal upside. But generative AI may start to erode these switching costs. Code generation, agentic systems, and automation are lowering the cost — and increasing the confidence — of migrations that once seemed harrowing.

Founders building infrastructure today need to design for this moment: if you’re replacing a legacy system, your timing may finally be right. If you’re defending an incumbent position, be intellectually honest about what’s actually keeping customers loyal.

3. Legacy systems endure for more than one reason

Not every legacy system is a house of cards. Some are technological marvels in their own right, and can still outperform modern systems under specific conditions.

“The mainframe is an I/O supercomputer,” Jaffe explained. “If you’re trying to build a global-scale credit card processing system or ATM network… you’d be hard-pressed to find a system that can run tens of thousands of concurrent transactions against the same refrigerator-sized machine and have perfect ACID compliance.”

The implication is clear: Disruption is not just about being cheaper or more user-friendly. It’s about matching, if not exceeding, mature systems on reliability, scale, and precision. If your product can’t meet those standards, it’s not a replacement. It’s a supplement.

4. Elasticity of demand can unlock category expansion

In most markets, falling prices mean shrinking margins. In infrastructure, falling costs often lead to market expansion.

Jaffe points to the historical drop in storage prices — from “$500,000 a gigabyte in 1982 to three or four cents today” — as a turning point. Rather than compressing the market, it created new use cases and vastly expanded demand.

That elasticity still holds. Founders building data infrastructure, vector search, or real-time analytics should consider not just the current TAM but also the potential TAM unlocked by cost compression and usability breakthroughs.

In short: don’t just ride the wave; predict where demand expands when you lower costs.

5. GenAI is redefining product boundaries and user expectations

One of the most striking shifts Jaffe sees is how genAI blurs traditional product categories. Data catalogs, ETL tools, even databases are being reimagined as “tools for the model,” not just tools for the user.

“You may see even database choices change,” he said. “Instead of doing search, chunking results, and passing them into a model… you give your model access to your data infrastructure as a tool.”

Founders building software for technical users must now ask: What if the primary user is an LLM? What interfaces, APIs, and workflows would that require? What kind of product becomes possible when models orchestrate tasks instead of humans?

*Note: Insight has invested in Precisely.