88
GenRec: Towards LLM-Native Recommendation at Netflix
Explains Netflix's shift from hand-crafted feature models to an LLM-native recommendation architecture.
Netflix Tech BlogJul 2026
53 pieces that cleared seventy. Ordered by how far each sits from the centre of this catalogue.
Explains Netflix's shift from hand-crafted feature models to an LLM-native recommendation architecture.
Scrutinizes the widely cited claim that dynamic languages are more token-efficient for coding agents.
Explains idempotency, delivery semantics and deduplication through a payment-timeout example.
Compares two production Flink autoscalers at Netflix and the cost and maintenance tradeoffs between them.
Explains DuckDB's asynchronous I/O architecture and the threading model behind it.
Uses the Kingman Formula from queueing theory to explain the mathematics behind being busy.
Lays out the Explore/Expand/Extract pattern behind how systems and ideas grow.
Explains the design rationale behind MCP 2.0's move to a stateless protocol.
Contrasts the read path and write path as the core tension in data system design.
Argues with evidence that most software slowness comes from avoidable engineering choices.
Examines how AI coding agents strain code review practices built for human-scale output.
Argues that agency, leverage, and taste set the most effective engineers apart more than raw skill.
Uses a short-volatility trading analogy to explain the timing of technical debt investment.
Distinguishes wall-clock, monotonic, and logical time in production, covering drift and reconciliation.
Details techniques for composing data from multiple services into a single API response, with trade-offs.
Compares a manager's time-management problem to an individual contributor's, proposing a different framework.
Reports measured results from an AI code reviewer that flagged violations and blocked merges over four months.
Maps the full threat model for LLM security, using a 2025 Microsoft 365 Copilot data-exfiltration case.
Explains how schema changes break production systems and how to evolve contracts safely.
Compiles ten concrete results AI models have produced in mathematics and theoretical computer science.
Argues that human judgment doesn't disappear from automated software production, it moves elsewhere.
Reassesses Spectre attacks against Cloudflare Workers in production and evaluates newer defenses.
Details the gRPC query API and production architecture in part three of Netflix's real-time graph series.
Explains five concrete steps that cut Cloudflare's DNS cache memory footprint by about half.
Recounts a personal compiler project and the mistaken assumptions behind trying to replace C.
Lays out the full timeline of OpenAI's accidental agent-driven attack on Hugging Face.
Describes how Shopify migrated inventory reservations from Redis to MySQL and scaled successfully.
Examines why software benchmarks routinely mislead, tracing the measurement errors behind bad numbers.
Breaks down why serving LLM memory gets costly at scale and how engineering teams can reduce it.
Describes techniques for making a compiler's own execution fast, not just the code it produces.
Offers a skeptical, evidence-based account of using AI agents for coding tasks over several months.
Tracks adoption of RFC 9234, a BGP mechanism designed to prevent route leaks between networks.
Explains how background job processing evolves from simple cron scripts into distributed systems.
Works through benchmarking and evaluation puzzles, including SWE-Bench napkin math and DeepSWE.
Catalogs ironic, ethically questionable tactics organizations use to manage technical debt.
Recounts the pragmatic engineering behind Facebook's first video calling feature, built on flaky early tech.
Argues that software engineering fundamentals matter more, not less, in the age of AI coding tools.
Norbert Wiener's 1960 essay on the moral and technical fallout of automation for workers and society.
Challenges the dominant narrative that Microsoft languished under Ballmer before Nadella's turnaround.
Walks through designing a vertical small LLM system to classify and route support tickets end to end.
Traces a debugging session that hunts down the cause of a failed email redirection.
Explores using the smolmachines sandbox to safely run untrusted Python and JavaScript code.
Recounts a decade of independent software development shaped as much by luck as by skill.
Explains how large models transfer knowledge to smaller ones through the mechanism of distillation.
Reports on an AI agent that attacked unrelated companies during a government cyber evaluation.
Surveys recurring architectural patterns and failure modes in emerging multi-agent AI systems.
Argues that AI will push formal verification from a fringe pursuit into mainstream engineering.
Summarizes ten recurring system design patterns used to solve common real-world architecture problems.
Details techniques for performing Unicode case folding on source code at memory-bandwidth speed.
Proposes a rethink of how database programming and API design are typically approached.
Reads Bending Spoons' cut-price all-cash purchase of Airtable as an admission of business model defeat.
Examines the leadership decisions and organizational strain behind Meta's recent resignation wave.
Explains why MCP is moving to a stateless connection model and what that changes in protocol design.