# What Happens When Engineers Stop Reading Their Own Code?

## Executive summary

The discussion explores the evolution of software development into 'AI software factories,' where AI agents automate core engineering tasks like model routing, code review, and issue management. Key architectural patterns include the AI Gateway, which manages multi-cloud, multi-regional model calls and aggregates rate limits. Furthermore, advanced agent stacks (like Cloudflare's) are being built on frameworks such as Flue, enabling sophisticated, context-aware code reviewers and internal office suites (Cloudflare OS). The trend suggests a shift in engineering value from writing code to understanding high-level concepts and system architecture (the 'scholarly engineer').

## Key takeaways

- AI Gateways manage complex model routing and rate limits.: AI Gateways (used by Miro) are necessary to handle model calling across multiple regions and clouds (e.g., AWS, Azure, GCP, Anthropic Direct). They aggregate rate limits and define business logic, such as enforcing EU-only models for specific customers, making the model call layer highly self-contained and routable.
- Agent-powered code review focuses on human behavior, not just code fixes.: Advanced internal AI code reviewers (built on Flue) analyze not only suggested fixes but also the historical data on how many Pull Requests (PRs) are opened, merged, and closed. This feedback loop helps the system understand what is 'important to people,' moving beyond simple syntax checks.
- Agent stacks are becoming full-featured, open-source office suites.: Cloudflare developed Cloudflare OS, an open-source agent office suite that allows users to build applications, dashboards, and blueprints. This system is built on the Agents SDK and Flue, enabling agents to interact with specific applications within the internal network.
- The value shifts to conceptual understanding ('Scholarly Engineer').: As LLMs generate vast amounts of code, the skill set is moving up the abstraction layer. The 'scholarly engineer' is defined as someone who understands the underlying concepts and ideas of an industry, rather than needing to read every line of code.

## Technical details

- AI Gateway Architecture: A centralized layer responsible for model calling, handling self-hosted and external LLM calls, and managing complex routing decisions across multi-regional and multi-cloud providers (e.g., AWS, Azure, GCP).
- Agent Frameworks and Control Flow: Flue is a JavaScript framework (from the makers of Astro) designed for building agent control flow. It allows applications to deploy anywhere JavaScript runs (Node.js, Vercel, Cloudflare Workers), providing advanced capabilities like lazy loading of tools.
- Code Review Automation: Miro uses Easel bot, which reads production logs overnight to proactively open fix PRs. Cloudflare uses an internal AI reviewer built on Flue that analyzes PRs based on company context and human approval/rejection patterns.
- Security and Compliance: Oplane is a security check tool for PRs that provides a comprehensive review of compliance, secret exposure, and data redaction recommendations, ensuring merges are secure.

## Practical implications

- Build engineers should focus on building robust, multi-layered abstraction layers (like AI Gateways) to manage external dependencies (LLMs, cloud providers) and complex business logic.
- The concept of 'Agent Experience' is a new discipline; understanding how agents interact with internal tools and data is critical for future product design.
- Implementing internal 'Codex' systems that capture company-specific best practices and operational context is essential for maximizing the effectiveness of internal AI agents.
- Automating the code review process requires incorporating human feedback loops (e.g., PR merge/reject statistics) to continuously refine the AI's understanding of 'value.'

## Topics

AI Engineering, DevOps, Agentic Workflows, Open Source Development, Software Architecture, Cloudflare, Miro, Flue, Astro, Cloudflare OS, Oplane

Source: https://www.youtube.com/watch?v=Rmtq-CRex4M
