Mike Spitz is chief technology officer of Pro Football Focus (PFF), the sports-data company serving professional and college football teams and consumers. He is reshaping software delivery around a specific premise: when AI agents accelerate implementation, engineering organizations must optimize specifications, coordination, testing, and customer feedback instead of individual coding speed.
Spitz initially tested coding agents on low-risk projects before recruiting two experienced engineers for a three-month experiment inside PFF’s distributed engineering organization. Their agent-assisted team deployed more frequently than a larger conventional team, although Spitz acknowledges that smaller teams naturally move faster. His decisive measure was whether the resulting software improved customer satisfaction—not pull-request counts or code volume.
His account of PFF’s agent-first engineering experiment describes a post-engineer engineering organization that retains engineers while concentrating their judgment on architecture, security, product quality, and clearly defined requirements.
- Optimize the organization for agent speed. Replace standups, sprint planning, and manual status updates with automated ticket coordination, short cross-functional huddles, rapid releases, and immediate production feedback.
- Make design intent executable. Human-defined specifications become agent-generated lightweight design documents reflecting established engineering patterns; engineers review them before agents produce sequenced tickets and pull requests.
- Encode engineering culture as composable skills. Capture team-specific conventions—including trunk-based development, feature flags, API architecture, analytics, and product presentation—so automated implementation preserves the organization’s standards.
- Judge delivery by customer outcomes. Faster deployments matter only when customers receive better products; satisfaction surveys and production feedback provide stronger signals than code volume alone.
- Automate verification before remediation. PFF’s acceptance-criteria-driven QA checks staging deployments against ticket requirements. Agent-generated fixes and self-healing workflows remain goals, not established capabilities.
Spitz advocates introducing these practices gradually: begin with senior engineers, repetitive work, low-risk systems, and reliable guardrails before expanding agent autonomy.