The Pragmatic Engineer — searchable YouTube transcripts & AI Q&A

The revised rules of engineering management with Will Larson

Will Larson argues that the 2010s management ethos of never making technical decisions and prioritizing engineers' personal interests is outdated, and that today's engineering leaders must be deeply involved in technical details and business problems. He recommends that middle managers reskill by getting back into the details of software development, even if it means taking a lateral or backward step, as the industry is changing rapidly with AI. He also emphasizes that durable, high-ownership teams are more effective than constantly rotating people to crises.

Uber’s AI infra, from the inside

Uber's AI infrastructure is built on a foundation of pre-existing developer tools, making it easier to retrofit AI capabilities, argues Udai Kira Mendesi. The company uses a model gateway and MCP gateway to abstract models and tools, enabling agents to access internal services and vendor tools uniformly. A key takeaway is that production-ready AI code requires more than just a coding harness; it needs robust infrastructure for testing, validation, and review, such as Slate for environment routing and Capri for capture-replay testing.

Building resilient systems with Sam Newman

Sam Newman argues that microservices should be an architecture of last resort, adopted mainly to achieve organizational autonomy, not as a magic solution. He emphasizes that most outages stem from resource saturation, one of his three rules of distributed systems, and that resilience requires defining what "good" looks like through observability and SLOs. He also cautions against over-relying on AI, recommending modular architectures and human oversight to avoid cognitive surrender.

The state of the tech industry in 2026 - LDX3 New York

The speaker argues that the tech industry has undergone a rapid transformation in the past year, with AI-generated code becoming the norm and engineers increasingly working with multiple agents in parallel, leading to smaller teams and the decline of traditional IDEs. However, he notes that teams remain the unit of work, planning and testing are still crucial, and non-engineers are not yet shipping production code. He also highlights challenges such as code quality issues, code review theater, and the strain on engineering leadership, while predicting a future of cloud-based agents, refactoring waves, and a greater emphasis on domain knowledge and AI fluency.

Distributed databases with Peter Mattis

Peter Mattis argues that AI coding tools amplify the expertise of domain experts, making them dramatically more productive while maintaining high code quality. He describes how he used AI to implement a 10,000-line B-tree in Rust in about 30 minutes, and emphasizes that the key is to guide the AI with deep knowledge of the system, including testing and performance considerations. He also notes that AI agents tend to be lazy about testing, but this can be corrected with firm guidance and advanced testing techniques.