Focus on creating specialized AI agents that mimic real employees with specific roles, like a recruiter named "Jim." This approach fosters better integration and effectiveness by making the AI behave, look, and feel like a real coworker.
AI tools now handle much of the data ingestion and synthesis that PMs traditionally managed. The modern PM’s value lies in using this AI-surfaced customer context to make high-leverage trade-off decisions and bridge the gap with go-to-market teams.
Instead of focusing on managing people, leaders should spend significant time reviewing the actual work. This direct engagement creates a "case law" system where top-level decisions on quality trickle down and establish the standard for the entire organization.
Relying solely on tool constraints within an agent is insufficient, as it can always find ways to cross the network boundary. A more robust approach is to use a proxy that intercepts all network requests, allowing for policy enforcement at the network level.
Instead of manually writing complex security policies for agents, run the agent in an audit mode to capture its network traffic. Then, use an LLM to analyze this traffic and automatically suggest a comprehensive, baseline security policy based on observed behavior.
As companies adopt AI, token costs become a significant, opaque expense. This creates a need for new financial tools that provide granular analytics to track, understand, and optimize AI spend across models, use cases, employees, and even per customer.
An engineer’s ability to leverage AI tools is a critical measure of their productivity. Brex has shifted its interview process to practical building exercises that are too complex to complete without proficient use of AI, directly testing for 10x potential.
The traditional SaaS model provides customers with a tool to do a job. With agentic AI, companies can now sell the completed job as a service. This represents a fundamental shift in business models, where the value is the direct outcome, not the software.
To prevent building great solutions for the wrong problems, structure reviews into two phases. First, a "Problem Alignment" meeting uses data to agree on what to solve. Only then does a "Solution Alignment" meeting review the proposed fix, ensuring efforts are correctly focused.
To foster a culture of AI-driven productivity, don't throttle usage with cost controls initially. Let employees experiment deeply to discover high-leverage use cases. Once adoption is widespread, introduce analytics to surgically optimize low-ROI spending without stifling innovation.