Agents Don't Do Standups: Building the Post-Engineer Engineering Org — Mike Spitz, PFF

Key Concepts

• Agentic Engineering: Utilizing AI agents to automate the software development lifecycle (SDLC) rather than just using AI as a coding assistant.
• Lightweight Design Document (LDD): A concise, AI-generated technical specification that ensures code consistency and architectural alignment.
• Trunk-Based Development: A version control management practice where developers merge small, frequent updates to a core "trunk" or main branch.
• Composable Skills: Breaking down engineering tasks into modular, repeatable units (e.g., feature flag creation, API generation) that AI agents can execute.
• Self-Healing Pipelines: An automated QA process where agents identify failures in acceptance criteria and automatically generate PRs to fix them.

---

1. Case Study Overview: PFF Engineering Transformation

PFF, a sports data company, transitioned its engineering workflow between January and March to address competitive stagnation. Despite having 20 engineers, the team was falling behind. By integrating AI agents into their workflow, they achieved:

• Deployment Frequency: Increased to 5 deploys per day (compared to 1 every 5 days for the larger team).
• Output Efficiency: A 10x increase in output, measured by blending ticket volume with code complexity.
• Project Velocity: Features that previously required four months were completed in under two months.
• Customer Satisfaction: Average quality scores rose from 7.5/10 to 8.6/10.

2. Methodology: The "Factory" Framework

The speaker advocates viewing the SDLC as a factory floor, breaking down complex tasks into small, composable elements.

• The Workflow:
  1. Spec & Interview: The agent interviews the engineer to define the requirement.
  2. LDD Generation: The agent creates an LDD based on historical architectural patterns.
  3. Distribution: The LDD is shared for peer feedback.
  4. Automation: Tickets and PRs are automatically generated based on the LDD.
  5. QA: Upon merging to staging, a QA agent validates the code against the original acceptance criteria.

3. Process Changes and "De-Scrumming"

The team eliminated traditional Agile ceremonies that were deemed redundant in an agent-driven environment:

• No Sprint Planning: Estimations are unnecessary when agents handle the bulk of the execution.
• No Daily Standups: Status updates are automated; tickets update automatically based on PR status (In Progress -> Review -> Merged).
• Huddles: Replaced formal meetings with 30–60 minute "huddles" every other day for rapid feedback and alignment.
• Retrospectives: Replaced by direct customer satisfaction surveys and hard development metrics (deployment frequency).

4. Strategic Implementation Advice

• Start with "Boring" Tasks: Focus on repetitive, low-risk tasks that engineers dislike (e.g., style linting, variable naming, boilerplate code).
• Select the Right Talent: Start with the most senior engineers who possess deep system knowledge. The ideal "AI-era" engineer is curious and capable of managing high-level architecture rather than just writing code.
• Phased Approach: Avoid the "hackathon" trap of giving everyone access at once. Start with non-critical systems to build trust in the agents.
• Guardrails: Ensure security and architectural patterns are encoded into the agents' "skills" before scaling.

5. Key Arguments and Perspectives

• Engineers are no longer the bottleneck: The bottleneck has shifted to the speed of product definition and the ability to maintain a consistent "brand feel."
• The "Sports Car" Analogy: Not every engineer will thrive in this new environment. Those who rely on highly prescriptive, manual specs may struggle, while curious engineers who understand the "why" behind the code will excel.
• Compounding Advantage: The speaker warns that being "a few months behind" in adopting these workflows can quickly spiral into being a year behind due to the compounding nature of AI-driven development speed.

6. Notable Quotes

• "Instead of figuring out how we can help engineers go and output more, how do we help make the agents quicker?"
• "Not everyone can drive a sports car. And that’s all right."
• "What is the purpose of this meeting? Is it just because everyone else has been doing it before, or is it because it actually helps out?"

Synthesis

The PFF case study demonstrates that the future of engineering lies in Agentic SDLCs. By treating the development process as a modular factory, companies can move away from heavy, process-laden methodologies like Scrum toward a high-velocity, autonomous model. The primary takeaway is that success depends on encoding organizational patterns into AI skills and focusing human effort on high-level design, product feel, and customer satisfaction rather than manual coding and administrative overhead.