Arvinth Shriram

I know the wickets offer so little to bowlers but the gulf between batting skills and bowling is now huge. And threatens to get even bigger

Anthropic’s agents won’t kill SaaS. But something else will. The “SaaSpocalypse” narrative is everywhere right now. When Anthropic dropped Claude Cowork plugins in Feb, $285 billion+ evaporated from software valuations in 48 hours. I’ll be the first to admit: I have skin in the game. As an early-stage AI investor and founder of Freshworks, my perspective is naturally colored by that journey. But looking past the hype, here is my honest take on where we are actually headed. 1. Will “Vibe Coding” Kill SaaS?
 There’s a popular idea that because anyone can now “vibe code” an app in minutes with Claude Code, the value of SaaS drops to zero. First, vibe coding is not a SaaS death sentence. Yes, you can now describe your requirements in plain English and Claude/Cursor spits out a working app in minutes. I’ve seen solo founders replace $300/mo tools overnight. Cool. But that’s a weekend prototype — not a system that survives 10k concurrent users, SOC2 audits, 17 legacy integrations, and the 3 a.m. pager storm. But the most important point is you are vibe-coding yesterday’s legacy systems based on structured forms. The right approach should be to start reimagining what software should feel like when AI is the operating system. 2. Agents vs. Systems of Record 
The “Claude agents will kill SaaS” argument misses a fundamental architectural reality. If an agent sits on top of HubSpot, Salesforce, or Freshworks to deliver intelligence, how exactly does that kill the underlying platform? These applications are the System of Record. They are the crucial data input layer. To be effective, agents need:
• Structured data from CRMs and ERPs
• Unstructured data from emails, Zoom summaries, and call recordings The agent is the brain. The SaaS platform is the memory and the nervous system. One cannot function effectively without the other. (Aaron Levie nailed it recently: agents will be SaaS’s biggest users.) There is another theory that you don’t need the underlying system of record anymore. Just connect the zoom meetings, emails and call transcripts and AI can figure out everything by itself - thats a very flawed assumption and we will do a separate post on why and how that will fail 3. So, Who Will Kill SaaS?
 SaaS isn’t going to be killed by a specific LLM release. It will be disrupted by whoever reimagines the solution from the ground up. Adding an AI layer on top of legacy architecture is a stop-gap—a “wrapper” strategy. The real killer will be either:
• An incumbent brave enough to disrupt its own legacy architecture, or
• An AI-native startup that builds a system where AI isn’t an “add-on,” but the core foundation of the logic. The Path Forward
We are moving from a world of “software you use” to “software that does.” The transition won’t be overnight, but the blueprint is changing fast. In my next post, I’ll outline exactly what a reimagined system of enterprise software—built from day one for the AI era—actually looks like.