# Coding Agents Don't Scale Themselves. Neither Do Your Teams. — Patrick Debois, Tessl

## Executive summary

The shift toward autonomous systems (the 'dark factory') is not limited by technology but by organizational readiness. The core message is that the focus must move from fixing code produced by AI agents to improving the underlying system and processes. Scaling requires moving beyond solo developer efforts to establishing centralized, reusable platforms and mandates for context authoring and tooling.

## Key takeaways

- Organizational Readiness vs. Technology Limits: The resistance to advanced automation (like continuous delivery or the dark factory) is not due to technological impossibility but because organizations are not yet structured for it. The differentiator will be the team, platform, and organization, not the technology itself.
- Shift Focus from Code Fixing to System Improvement: Developers should stop focusing on fixing the code produced by agents. Instead, they must improve the system architecture and processes (e.g., improving test coverage or documentation generation) that guide the agent.
- Scaling Requires Platform Ownership: To scale automation beyond individual teams, organizations must establish centralized 'paved roads' and dedicated owners for reusable components (e.g., authentication systems, linters, or context registries). This prevents technical sprawl.
- Key Metrics for Measuring Progress: Productivity should be measured by two metrics: the reduction in 'human touches' required to get a correct result, and the degree of fix/improvement that is shared across multiple users (the multiplier effect).

## Technical details

- Agent Enablement Patterns: The process involves moving through stages: Prompting $\rightarrow$ Specs/Context Engineering $\rightarrow$ Harnesses and Loops. The goal is to elevate practices from simple auto-completion to system thinking, instructing agents with full engineering practices (e.g., 'Please do it with tests. Please update the documentation.')
- Systemic Planning and Retrospectives: Planning sessions should split into two types of work: those sufficiently well-defined for agents to pick up, and conversational items requiring human decision-making. Retrospectives must shift from discussing code issues to identifying systemic weaknesses the agent repeatedly hit.
- Platform Requirements: The platform team needs to evolve its focus beyond traditional infrastructure (like MCP gateways) to include managing skill registries, evaluation systems for context/guardrails specific to coding agents, and identity management.

## Practical implications

- Mandate team leads and platform teams to drive automation efforts rather than relying on individual developer initiative.
- When hiring for AI roles, evaluate candidates not just on technical skill (AI/ML) but also on their ability to share knowledge and collaborate within a system.
- To manage costs, focus on optimizing the process by improving context and harnesses, rather than simply limiting all spending.

## Topics

Dark Factory, Continuous Delivery, AI Agents, Platform Engineering, DevOps Maturity, Tessl, Patrick Debois' Website/Patterns

Source: https://www.youtube.com/watch?v=zCJtYuqwm7E