The future of software
A crucial debate on disruption and durability
The future of software
A crucial debate on disruption and durability
A hugely impactful session brought together Marcus Ryu (Battery Ventures), Russell Kaplan (Cognition AI), and Alex Kayyal (Lightspeed Venture Partners) for a frank discussion about software's future - with focus on how product will evolve, and the competition brought by native AI. Their debate crystallised the central tension facing every software company: Is AI an existential threat or an unprecedented opportunity?
The capability revolution
All participants acknowledge the ‘killer app’ of AI for coding and software development. Russell Kaplan offered an explanation of why - in coding, AI performance has dramatically outpaced other domains, due to clean feedback loops and existing quality controls (and perhaps, also the natural interests of those developing these tools).
Kaplan explained how AI transforms the entire development lifecycle, not just code generation. The implications are profound. For one Devin customer, AI handles 10,000 hours of engineering work monthly through automated testing and code quality improvements. Kaplan added that AI doesn’t necessarily replace coders, but enable tasks that "wouldn't have gotten around to in the first place" without AI.
"Most of the grunt work we see people having Devin do. It's not work that's replacing the grunt work that humans are doing. It's work that they just wouldn't have gotten around to in the first place unless you had AI."
The realist's perspective
Marcus Ryu, drawing from his experience building Guidewire to (today) a $20 billion market cap, offered a counterpoint to AI disruption. "I'm not an AI sceptic... I'd like to say I'm an AI empiricist," he stated, before systematically addressing the disruption narrative.
One key insight: switching costs and implementation complexity remain formidable. Even after 25 years, 80% of financial transactions still rely on COBOL code. "These systems were completely depreciated. The actual cost per transaction was de minimis," Ryu explained, illustrating why legacy systems persist despite alternatives.
For Ryu, the real question isn't technological capability but market dynamics:
It would only be an existential threat if the core value proposition of the software offerings... were no longer as apposite because the market no longer wanted it.
"You're going to have to make like 2 or 3 architectural decisions, and you better get them right. Because you can change your mind later, but it'll cost you. But get those right, and if you are unsure, then find people who will help you be more sure. But get those few decisions really right."
The switching cost battlefield
Kaplan agreed that if companies are surviving mainly due to high switching costs, then they do, in fact, face existential risk as AI makes migrations trivial. "The first really good application of coding AI agents is actually migrations and refactors." Database migrations, historically painful enough to lock in customers, become straightforward with AI assistance. Infrastructure software vendors who've coasted on technical switching costs may find their advantages evaporating.
The bull case for transformation
Alex Kayyal provided the synthesis, framing the current moment as "SaaS 4.0." His perspective: software will increasingly "bend to the needs of the user" rather than forcing humans to adapt to limitations. The opportunity isn't just efficiency but fundamental reimagination. "Oracle stock is up 250%, Salesforce is up 50% over five years," Kayyal noted, suggesting that companies embracing AI infrastructure, strengthen their positions. The key is abstracting away the "unit of value" - the work to be done - rather than just providing tools.
"Historically, humans had to sort of bend around the limitations of the software. There was a tool out there that companies would create. And then humans would sort of find workarounds based on what the gaps were... I think that's about to get flipped on its head where software will bend to the needs of the user."
Hard-won wisdom
The panel converged on several crucial insights for incumbents:
1. Domain expertise matters more than ever
Understanding workflows, compliance, and customer needs creates value even as technical competitive advantage erode
2. Speed of experimentation trumps perfect architecture
Kaplan advocated borrowing from ML research practices: "Never go to bed with the GPUs idling"
3. Some decisions can't be reversed cheaply
Ryu emphasised that while experimentation is crucial, core architectural choices require careful consideration
The verdict
Rather than a binary outcome, the panel revealed a nuanced future. Companies with genuine customer value, domain expertise, and the agility to adapt will thrive. Those relying solely on technical lock-in or distribution advantages face real threats.
As Kayyal summarised: "In this new world, it's actually software will completely abstract away the unit of value which is the work to be done."
The winners will be those who successfully navigate this transition, combining incumbent advantages with startup-level agility.
