First impressions of GPT-5.5 from Will Koh

The Ramp demo is specifically about tool selection intelligence, meaning GPT-5.5 is being evaluated not just for raw reasoning but for its ability to route between functions reliably enough to trust in production fintech environments. That's a narrower and more demanding bar than general capability demos.

OpenAI is running a coordinated first-impressions series on the same day across very different verticals. The Claire Vo piece covered AI-native product builders, the Aaron Friel piece covered internal OpenAI workflows, and now Ramp represents regulated B2B finance. Taken together with the NVIDIA partnership announcement, which framed GPT-5.5 as a task-execution model rather than a chat model, the Ramp demo fits a deliberate pattern: OpenAI is building a case that the model holds up under domain-specific, high-stakes function-calling conditions, not just general benchmarks. The workspace agents rollout covered separately on April 24th adds further context, since autonomous tool routing is exactly the capability class those agents depend on.

Watch whether Ramp moves from demo to disclosed production deployment within the next two quarters. A public case study with measurable error-rate or latency figures would confirm that GPT-5.5's tool-selection gains are real in fintech workflows rather than curated for the launch window.