Why PFF Dropped Sprints and Stand-ups for AI

Mike Spitz, CTO at Pro Football Focus, replaced sprint ceremonies with async specs and demo huddles once AI agents took on the build work. A feature once scoped at two sprints shipped in a day and a half, and design became the new bottleneck.

Nov 5, 2026 27:38

Mike Spitz

Expert Insights

Mike Spitz is CTO at Pro Football Focus, where he leads engineering for the consumer side of the business: the fantasy and betting products built on PFF’s NFL and NCAA player data. He describes the company’s AI transformation as an existential necessity for that business. When it started, most of his engineers either avoided AI or used it without getting much out of it, and parts of management saw no point in it. Instead of mandating it, Mike won people over by automating the chores they disliked most.

Hear him outline:

Why buy-in came from automating chores engineers kept flagging in PRs, like feature flags and analytics, and not from mandates
How a shared spec skill and a lightweight design document give Claude enough context to break features into tickets
Which work runs fully autonomously for hours or days, and why anything that needs taste stays with people
Why the team builds its own skills, shares them as a Git submodule and clears context at 30–40%
Why AI-generated design gets PFF only 60–70% of the way, and the risk of an app that feels like every other app
How sprint planning, refinement and stand-ups gave way to async specs, demo huddles and fast customer feedback

In an environment where creating apps is really easy, do you want your app to feel like every other app or do you want it to feel special?

— Mike Spitz, CTO at Pro Football Focus

Monterail Team Analysis

Build the adoption backlog from friction. Recurring PR comments like “we forgot the feature flag” are a ready-made list of automation targets. Each fix takes minutes as a skill and wins over the engineers who were skeptical.
Gate autonomy on automated validation. PFF only gives the model multi-day batches of tickets in areas with automated checks. Before you scale agent work, invest in tests and checks for the code you plan to delegate.
Expect the bottleneck to move upstream. The Tiger Team’s backlog of pre-designed, pre-specced work ran out within a few months. If design and spec capacity don’t grow with engineering speed, faster engineering just means engineers waiting.
Audit ceremonies by the information they move. Stand-ups and refinement existed to route information between people. Once specs and a knowledge base carry that load, most of those meetings lose their purpose. Demos survived because they need human judgment in the room.
Measure customer satisfaction, not output. Twenty thousand survey responses give PFF a quality signal that PR counts and lines of code can’t provide. The satisfaction scores for features built after the shift are the strongest evidence that the transformation is working.
Keep humans on core UX when differentiation matters. Models trained on the average produce average interfaces. Teams with proprietary data or a distinct brand risk losing that edge if AI owns the cornerstone screens.
Plan non-engineering rollouts differently. For engineers, adopting AI protects their jobs. Support, finance or marketing staff may see it as a threat to theirs, and an outside “forward-deployed” engineer won’t have the domain context. Involve the people who own those roles from the start.