Keep pulling the thread on Pedro Franceschi.
Pedro Franceschi believes a company's CEO must also serve as its chief AI officer, taking personal responsibility for understanding the technology's boundaries.
Brex developed and open-sourced a tool called Crabtrap to secure its AI agents in production.
Pedro Franceschi advises founders starting a company today to begin with the premise of a one-person company and use AI to augment their capabilities.
At Brex, the company is building a 'customer world model' by ingesting every customer touchpoint, from dashboard clicks to support calls, to predict future customer needs.
Pedro Franceschi predicts that AI-related costs, such as token consumption, will easily become the single largest expense category for most companies.
Brex is redesigning its entire customer onboarding process to use AI for KYC checks on leads, not just customers, enabling earlier risk assessment in the sales funnel.
Pedro Franceschi advises CEOs of non-AI native companies to operate as if they are executing a turnaround, refounding the company's identity and processes around AI.
Brex has implemented a self-improving system where bugs flagged in conversations with its expense agent automatically trigger another agent to modify the codebase and prompts to pass the new evaluation case.
Pedro Franceschi considers the release of reasoning models like Opus 4.5 to be a technological invention as significant as the invention of electricity.
Brex's Crabtrap tool functions as an HTTP proxy that analyzes an AI agent's entire network boundary to create and enforce auditable security policies.
At Brex, 98% of AI agent requests are automatically approved by its Crabtrap security policy, while the remaining 2% are evaluated by an LLM judge.
Pedro Franceschi believes the key capability that made Brex's security team comfortable with production AI agents was using an LLM as a judge for network requests.