Org design is the 'hidden architecture' of engineering
What happened
A recent talk on engineering leadership argues that an organization's chart is its architectural diagram, directly shaping technical and product outcomes. The discussion highlights the need to balance team autonomy with strategic alignment, particularly for platform teams that must be embedded in user feedback loops. For staff-plus engineers, impact is defined by demonstrating how systems fit together, not just by writing complex code.
Why it matters
- The concept discussed is a direct application of Conway's Law, first articulated by Melvin Conway in 1967, which states that organizations are constrained to produce designs that are copies of their own communication structures. This means a company with siloed teams will likely produce a disjointed, monolithic system. - For Staff-plus engineers, moving from a "Solver" to a "Finder" or "Driver" is key to increasing impact. This involves proactively identifying and orchestrating the resolution of high-impact issues that span multiple teams or product lines, rather than just solving problems assigned to them. - Platform teams can improve developer experience (DevEx) by treating their platform as a product, with internal developers as their customers. This involves creating tight feedback loops through surveys and direct communication, and having a dedicated platform product manager to own the roadmap and prioritize features based on developer needs. - To measure the success of platform initiatives and justify investments, teams use frameworks like DORA (Deployment Frequency, Lead Time, Mean Time to Recovery, Change Failure Rate) and SPACE (Satisfaction, Performance, Activity, Communication, Efficiency). Developer satisfaction is often tracked via Net Promoter Score (NPS) surveys. - In the context of AI, platform teams are evolving to become "AI enablers" for their organizations. This means providing governed, secure access to AI models and tools through internal APIs and "golden paths," thereby managing the risks of "shadow AI" adoption. - Large Language Models (LLMs) are being used to automate the generation of API documentation from OpenAPI specs, which helps to ensure consistency and accuracy, especially in microservices architectures. This treats documentation as a product that can be built and updated programmatically within CI/CD pipelines. - For API observability, machine learning is used for predictive insights and automated root cause analysis, moving beyond traditional monitoring. AI-driven tools can detect anomalies in API usage, forecast potential outages, and identify security threats in real-time. - When productizing AI capabilities, organizations often choose to build custom models, bolster existing products with third-party AI services, or buy off-the-shelf AI-powered tools for internal productivity. The decision depends on whether they have a unique data advantage and the desire to monetize the AI model itself.
Key numbers
- - The concept discussed is a direct application of Conway's Law, first articulated by Melvin Conway in 1967, which states that organizations are constrained to produce designs that are copies of their own communication structures.
What happens next
- This means a company with siloed teams will likely produce a disjointed, monolithic system.
Quick answers
What happened in Org design is the 'hidden architecture' of engineering?
A recent talk on engineering leadership argues that an organization's chart is its architectural diagram, directly shaping technical and product outcomes. The discussion highlights the need to balance team autonomy with strategic alignment, particularly for platform teams that must be embedded in user feedback loops. For staff-plus engineers, impact is defined by demonstrating how systems fit together, not just by writing complex code.
Why does Org design is the 'hidden architecture' of engineering matter?
The concept discussed is a direct application of Conway's Law, first articulated by Melvin Conway in 1967, which states that organizations are constrained to produce designs that are copies of their own communication structures. This means a company with siloed teams will likely produce a disjointed, monolithic system. For Staff-plus engineers, moving from a "Solver" to a "Finder" or "Driver" is key to increasing impact. This involves proactively identifying and orchestrating the resolution of high-impact issues that span multiple teams or product lines, rather than just solving problems assigned to them. Platform teams can improve developer experience (DevEx) by treating their platform as a product, with internal developers as their customers. This involves creating tight feedback loops through surveys and direct communication, and having a dedicated platform product manager to own the roadmap and prioritize features based on developer needs. To measure the success of platform initiatives and justify investments, teams use frameworks like DORA (Deployment Frequency, Lead Time, Mean Time to Recovery, Change Failure Rate) and SPACE (Satisfaction, Performance, Activity, Communication, Efficiency). Developer satisfaction is often tracked via Net Promoter Score (NPS) surveys. In the context of AI, platform teams are evolving to become "AI enablers" for their organizations. This means providing governed, secure access to AI models and tools through internal APIs and "golden paths," thereby managing the risks of "shadow AI" adoption. Large Language Models (LLMs) are being used to automate the generation of API documentation from OpenAPI specs, which helps to ensure consistency and accuracy, especially in microservices architectures. This treats documentation as a product that can be built and updated programmatically within CI/CD pipelines. For API observability, machine learning is used for predictive insights and automated root cause analysis, moving beyond traditional monitoring. AI-driven tools can detect anomalies in API usage, forecast potential outages, and identify security threats in real-time. When productizing AI capabilities, organizations often choose to build custom models, bolster existing products with third-party AI services, or buy off-the-shelf AI-powered tools for internal productivity. The decision depends on whether they have a unique data advantage and the desire to monetize the AI model itself.
