Ramp CTO launches Inspect, an AI coding agent built for enterprise complexity and multi-model flexibility

Rahul Sengottuvelu, Ramp's CTO, has launched Inspect, a coding agent built internally before being opened to external customers. The origin story matters: Ramp's engineers were using local coding agents and asking for bigger laptops, sometimes walking the office with them open. Sengottuvelu wanted to move that experience to the cloud, securely, and at the scale and complexity of Ramp's own codebase.

What Inspect does

The agent is multiplayer by design. Sessions are public by default, so anyone at the company can see what's happening, share a link, and continue a conversation with the agent. It integrates with Slack and works across all major model providers, with Ramp running its own inference for the product. The deeper technical lift was getting the agent to work inside Ramp's own systems, including databases, servers, and the full codebase, which Sengottuvelu describes as an "enormous engineering challenge."

“We just launched our coding tool — Inspect. The big focus for us was to make sure it works in a company of our scale, our speed, our complexity. We made it extremely collaborative, multiplayer first, works in Slack, every session is public by default. It works cross platform across all models. We also run our own inference for Inspect.”

The CFO angle

Sengottuvelu's pitch to finance leaders is spend visibility, not just developer productivity. Inspect pipes session-level data into Ramp's router, exposing which department used which model, at which service tier, with how much reasoning effort. CFOs get controls over AI coding spend alongside the engineers who are generating it. That framing is consistent with Ramp's broader positioning: Sengottuvelu argues Ramp is not incentivized to drive up token costs, unlike the labs themselves.

Multi-model as a competitive bet

The model-agnostic architecture is a deliberate hedge. The leading model provider changes every few weeks, Sengottuvelu says, and forcing customers to pick one provider means locking in a context and repo setup that becomes hard to move. Inspect lets different teams run different models simultaneously, with Ramp managing inference across all providers.

Internal infrastructure first

Inspect is also Ramp's internal platform. Automations for code review, production checks, and cybersecurity workflows all run on top of the Inspect API. Sengottuvelu describes it as a foundational layer, sitting on top of Ramp's router, with customer-facing products built above it. The implication is that Inspect is not just a product launch but the abstraction layer Ramp is building its own automation stack on.

The competitive field is crowded, with cursor, Windsurf, Claude Code, and Codex all vying for developer workflows. Ramp's differentiator, at least as Sengottuvelu frames it, is that none of those tools surface spend data to finance teams or run natively inside an enterprise's own systems at this level of integration.