Rocks2GA founder Ishan Mukherjee on building an agentic CRM and the pressure driving AI SaaS fraud

Summary

Ishan Mukherjee is building Rocks, which he describes as the first enterprise-ready agentic CRM. The product sits inside a customer's data warehouse — Snowflake being the clearest example — indexes go-to-market data that has drifted out of legacy CRM systems, and feeds a swarm of agents designed to make individual sales reps two to three times more productive today, with a longer-term path to ten times. Mukherjee notes that roughly 40% of data in a typical software warehouse is go-to-market data, and that traditional CRMs have lost both users and data to newer tooling and warehouses — which is the gap Rocks is trying to occupy.

The product surfaces through a web app, iOS app, Slack, and email. It runs its own warehouse-native CRM and syncs bidirectionally with Salesforce or HubSpot, so it is not trying to rip out existing systems of record so much as become the front-end layer sellers actually use. At MongoDB and similar enterprise accounts, Rocks integrates via API into internal tooling. The commercial motion is land-and-expand: start with top-performing individual reps, then grow to 200-plus active users across a customer like Ramp.

Centaur model, not full automation

Mukherjee is explicit that Rocks is building "bat suits, not butlers." Front-office AI error rates have to be near zero because a bad email to an enterprise prospect does not cost five minutes — it degrades brand trust and potentially loses revenue. The agents handle back-office work; the senior rep reviews and sends. He expects significant seat compression over time in the support roles around top sellers, but argues the near-term product has to be usable today rather than vaporware premised on autonomous agents that are not yet reliable enough for customer-facing work.

Incumbent threat

On the Salesforce question — Logan Bartlett's argument that the Salesforce of AI may just be Salesforce — Mukherjee's counter is that incumbents have lost the data as much as the users. Data has migrated to warehouses, usage has migrated to newer tools, and the winning position is to become the custodian of that warehouse-resident go-to-market data before Salesforce or ServiceNow reasserts itself. The bet is that distribution and legacy install base are less durable advantages when the underlying data has already moved.

Pressure and fraud risk in AI SaaS

Mukherjee acknowledges the benchmark shift that is creating pressure across the category. In 2021, best-in-class startups were reaching $1M ARR in nine months. The current expectation is $10M ARR in three months. He argues that chasing those curves at the expense of customer quality is a losing strategy, pointing to Datadog and Wiz as models — both were fast-growing but built on essential, sticky solutions for large enterprises rather than top-line velocity for its own sake. He does not name specific companies committing fraud but confirms the pressure is real and that some founders are misstating facts as a result.

The second source of pressure he flags is harder to manage: enterprise buyers have been exposed to ChatGPT and Perplexity and now expect software to behave like a fully autonomous agent. Delivering a probabilistic agentic experience to customers with Singularity-level expectations is, in his framing, the thing that actually keeps him up at night — more than the revenue benchmarks.

Rocks is pre-launch publicly, with a major announcement planned in the coming months. Current customers include Ramp, Redis, and MongoDB, all sourced through word of mouth.