Platforms for Humans and Machines: Engineering for the Age of Agents — Juan Herreros Elorza

Key Concepts

• Platform Engineering: The discipline of building and maintaining internal developer platforms (IDPs) to abstract infrastructure complexity.
• AI Agents: Autonomous or semi-autonomous software entities that use LLMs to perform tasks, write code, and interact with systems.
• Self-Service: A design philosophy where users (or agents) can provision resources without manual intervention or human-to-human requests.
• API-First Design: Building systems where all functionality is exposed via well-defined APIs, enabling programmatic access for automation.
• Shift-Left: The practice of performing testing, validation, and security checks as early as possible in the development lifecycle.
• DORA Metrics: A framework for measuring software delivery performance (Deployment Frequency, Lead Time for Changes, Change Failure Rate, Time to Restore Service).
• SPACE Framework: A methodology for measuring developer productivity (Satisfaction, Performance, Activity, Communication, Efficiency).

1. The Challenge: Human vs. Machine Workflows

The speaker, Juan Herrero Salorza (Team Lead at Banking Circle), highlights a common friction point in engineering: the "dependency loop." Developers often struggle with manual deployment processes, requiring them to consult teammates or infrastructure teams to resolve pipeline errors or provision dependencies (e.g., databases, blob storage).

While humans can navigate these social and technical hurdles, AI agents cannot. An agent cannot "walk to the second floor" to ask for help. Consequently, existing inefficiencies in platform design become "limiting factors" that prevent AI agents from achieving their full potential.

2. Framework for AI-Ready Platforms

To enable AI agents to be productive, platforms must evolve. The speaker proposes the following methodology:

• Self-Service & Automation: Remove human gatekeepers. If an agent needs a resource, it must be able to trigger the provisioning process autonomously.
• API-First Architecture: Agents excel at interacting with well-defined APIs. Platforms should provide discoverable APIs with schema validation and robust authentication/authorization, allowing agents to operate in a secure, iterative loop.
• Local-First & Shift-Left: Agents typically run locally. Platforms should allow agents to validate configurations and run tests locally before pushing to version control, reducing the feedback loop time.
• Observability for Machines: Traditional dashboards are useless to agents. Observability data (logs, metrics, traces) must be exposed via APIs or CLI tools so agents can verify if their actions were successful.
• Structured Documentation:
  • Centralized vs. Local: Keep documentation near the code for small projects; use a centralized, API-accessible repository for platform-wide documentation.
  • Agent-Specific Context: Utilize files like agents.md, cloud.md, or compiler_instructions.md to provide the agent with explicit "how-to" instructions, coding standards, and deployment conventions.
  • Skills: Codify recurring tasks into "skills" (markdown-based instructions) that agents can reference.

3. Encouraging Contributions and Guardrails

The speaker argues that AI lowers the barrier to entry for contributing to internal platforms. However, this requires a balance:

• Guardrails: Implement automated policies to ensure security and compliance.
• Standardization: Use the aforementioned agents.md files to ensure that even when developers (or their agents) contribute, they adhere to established organizational standards.

4. Measuring Success

To validate that platform improvements are effective, teams should track:

• DORA Metrics: To assess the impact on delivery speed and stability.
• Support Requests: A decrease in manual support tickets indicates that the self-service model is working.
• Developer Experience (DevEx): Use frameworks like SPACE to measure the qualitative impact on the team.

5. Notable Quotes

• "Best practices are still best practices... they are just much more obvious and perhaps much more painful now that we have these coding agents working next to us."
• "If it is technically self-service but it requires fetching some building blocks from five different places... then it's not really self-service."
• "Take advantage [of the AI hype]. You can use AI as the excuse to implement some best practices that... were always best practices if you didn't have the chance to do it until now."

Synthesis

The core takeaway is that AI agents are catalysts for long-overdue engineering discipline. By forcing platforms to be API-driven, self-service, and well-documented, organizations not only enable AI productivity but also resolve the technical debt and communication silos that have historically hindered human developers. The transition to an "AI-ready" platform is essentially a transition to a more mature, automated, and scalable engineering culture.