Agents Don't Do Standups: Building the Post-Engineer Engineering Org — Mike Spitz, PFF
Summary
Mike Spitz, CTO of PFF (Pro Football Focus), presents results from a three-month case study conducted January through March 2026, comparing two engineers using AI agents against a team of ten engineers working on the same codebase. PFF is a sports data company serving NFL and NCAA teams, with a consumer arm for fantasy football and sports betting that handles 100 million page views annually and 9 million drafts per year. The company had around 200 employees and 20 engineers and was falling behind competitors before beginning the experiment.
The central reframe driving the case study was shifting the optimization question from 'how do we help engineers go faster' to 'how do we make the agents faster.' The two-engineer team achieved a 25x increase in deployment frequency — shipping five times per day versus the ten-person team's cadence of one deploy every five days. Output measured by a blended metric of ticket count weighted by code complexity came out at 10x. Critically, customer satisfaction scores rose from an average of 7 to 7.5 out of 10 pre-AI to 8.6 out of 10 post-AI, validated by statistically significant surveys.
The new development workflow follows a Spec → Lightweight Design Document (LDD) → Auto-generated Tickets → PR sequence. The agent interviews the engineer to build the spec, then generates the LDD by analyzing how previous LDDs were structured, ensuring architectural consistency across the codebase. Tickets are automatically created from the LDD and structured so none are blocking each other, with dependencies flagged explicitly. Sprint planning was eliminated because velocity estimates are no longer meaningful when the bottleneck is not human capacity. Daily standups were replaced because tickets auto-update their status based on PR state: a PR opened sets the ticket to in-progress, entering review updates it, merging closes it. Sprint refinement and retrospectives were also dropped, with customer satisfaction surveys serving as the primary retrospective signal.
What survived is a biweekly half-hour huddle with engineers, product, and design to review what was built and collect rapid feedback, with a preference for deploying to production in MVP state as quickly as possible. Code review was split: agents handle style, naming conventions, and opinionated formatting feedback — removing the emotional friction from those conversations — while engineers focus exclusively on system design and architectural decisions.
PFF implemented a QA agent that triggers automatically after every merge to staging. The agent reads all tickets associated with the deployment, checks acceptance criteria, and reports pass/fail. Spitz described work underway to extend this into a self-healing loop where the QA agent's failure reports are fed back to a code-generation agent that automatically opens remediation PRs.
Spitz emphasizes that the engineering development lifecycle should be viewed as a factory, with each step broken into small composable skills: branch naming, feature flag creation for trunk-based development, API scaffolding using the service-repository pattern, analytics instrumentation, and interactive element generation. He cautions against importing skills from external sources that embed conflicting software design opinions.
On team selection, Spitz argues that not every engineer thrives in this setup. The profile that succeeds is curious and self-directed, comfortable exploring how systems work without prescriptive guidance. The prescriptive engineer who relies on detailed specifications is predicted to struggle. He advises starting with the engineers who have the deepest system knowledge, experimenting in non-critical systems first, and using a slow phased rollout rather than a company-wide hackathon-style launch. He notes that small teams of 20 engineers can execute this transition far more easily than enterprises with thousands of engineers, but warns that falling a few months behind now compounds into six or twelve months behind soon due to the exponential nature of the gains.
Key Takeaways
- •Reframe the optimization target: ask how to make agents faster, not how to make engineers faster — this single shift unlocks process redesign.
- •Two AI-augmented engineers outperformed a ten-person team 10x on ticket complexity and deployed 25x more frequently (5/day vs. 1/5 days).
- •Customer satisfaction improved from ~7/10 to 8.6/10, proving output quality rose alongside velocity.
- •Replace sprint planning and standups with auto-updating ticket state tied to PR lifecycle; tickets move through statuses without human intervention.
- •Structure the development flow as Spec → LDD → Auto-generated Tickets → PR to eliminate coordination overhead entirely.
- •Split code review: delegate style, naming, and opinionated feedback to agents; reserve human review for system design and architectural decisions.
- •Build a QA agent that checks acceptance criteria against tickets on every staging deploy; extend it toward self-healing by feeding failures back to a code-generation agent.
- •Model the engineering lifecycle as a factory of composable skills: one skill per discrete, deterministic task (branch naming, feature flags, API scaffolding, analytics).
- •Start the rollout with your strongest, most curious engineers on non-critical systems before expanding to high-traffic production services.
- •Avoid importing community skills that embed conflicting software design opinions — encode your own org's patterns explicitly.
- •Small teams have a structural advantage in this transition; the compounding gains mean a few months of delay now can translate to a year of competitive disadvantage.
Transcript
[music] >> All right. So, I'm in the slot before the snacks. I'm here to speak about the host engineering org and a case study we've been doing at PFF started in January and we finished it in March. PFF is a sports data company. We help NFL and NCAA teams figure out what they should be doing and we also have a consistent We also have a consumer arm which does fantasy football, sport betting and for those of you who know American football there's a draft happening end of the month and so allows p...
Speakers
Topics
Related Sessions
