Top CPOs reveal 5 hard truths about shipping revenue-generating AI agents

The Gist

• Agent roadmaps require rebuilding your entire product, causing widespread delays
• Half of AI users' time spent feeding context makes integration layers critical
• Agents lack identities, creating accountability challenges in CRM systems
• Customers will build unpredictable solutions on top of your agent infrastructure

• Building AI agents within products is more challenging than anticipated, requiring a complete rebuild of workflows in chat interfaces.
• AI agents must be prioritized based on their potential impact, starting with workflows that cannot break anything if they fail.
• Democratizing AI development across product teams, rather than centralizing it, is crucial for scalability and innovation.
• Context-building consumes significant user time, making setup efficiency a critical factor in AI product adoption.
• AI agents must balance autonomy with deterministic execution, especially in critical workflows like data recovery.
• Verification and accountability are emerging as major bottlenecks in deploying AI agents in enterprise environments.

• Prioritize AI agent workflows that minimize risk, starting with non-critical tasks before moving to production-impacting ones.
• Decentralize AI development across product teams to handle the volume and complexity of agent workflows.
• Invest in reducing setup time for AI features, as it directly impacts user adoption and perceived product value.
• Develop robust verification and accountability mechanisms for AI agents to ensure trust and reliability in enterprise environments.

• 300 to 800 Gong calls a day (Glean's sales team generates this volume, which their AI agent processes to propose Salesforce updates.)
• 180+ legal engineers (Harvey employs this number of ex-lawyers to build bespoke AI agents for law firms.)
• 60% of the Am Law 100 (Harvey has penetrated this percentage of top law firms, scaling against partner-pocket economics.)