Best of Code Review — December 2025

  1. 1
    Article
    Avatar of elevateElevate·22w

    My LLM coding workflow going into 2026

    A comprehensive guide to using LLM coding assistants effectively in 2026. Key practices include starting with detailed specifications before coding, breaking work into small iterative chunks, providing extensive context to the AI, choosing appropriate models for different tasks, maintaining human oversight through testing and code review, committing frequently for version control safety, customizing AI behavior with rules and examples, leveraging automation as quality gates, and treating AI as a force multiplier rather than replacement. The workflow emphasizes treating LLMs as junior pair programmers requiring guidance while maintaining developer accountability for all code produced.

  2. 2
    Article
    Avatar of simonwillisonSimon Willison·22w

    Your job is to deliver code you have proven to work

    Software engineers must deliver proven, working code rather than untested contributions. This requires both manual testing (seeing the code work yourself, documenting steps, testing edge cases) and automated testing (bundling tests with changes). With AI coding agents like Claude Code, developers should train these tools to prove their changes work through testing before submission. The human developer remains accountable for ensuring code quality and providing evidence that changes function correctly.

  3. 3
    Article
    Avatar of zaidesantonManager.dev·22w

    5 engineering dogmas it's time to retire

    Five common software engineering practices deserve reconsideration: relying heavily on third-party packages creates security and maintenance risks, mandatory code reviews slow teams down unnecessarily, 2-4 week sprints drain joy from development, overusing feature flags creates codebase complexity, and avoiding all code comments is an extreme position. Each practice has merit but shouldn't be treated as absolute dogma. Engineering managers should balance these principles with their team's specific context rather than following them blindly.

  4. 4
    Article
    Avatar of programmingdigestProgramming Digest·23w

    No code reviews by default

    Raycast's engineering team operates without mandatory code reviews, allowing engineers to push directly to the main branch and request reviews only when needed. This trust-based approach emerged from their early startup days and scaled with their distributed team. They prioritize rapid iteration through daily internal releases, dogfooding changes within 24 hours, and shipping public updates every two weeks. Code reviews are still used selectively for unfamiliar code areas, database migrations, or onboarding new team members. The team relies on post-commit notifications, quick video calls, and continuous integration to maintain quality while avoiding the bottlenecks of traditional pull request workflows.

  5. 5
    Article
    Avatar of engineerscodexEngineer’s Codex·21w

    Everyone is a Staff Engineer Now

    AI coding agents like Claude Code are shifting engineering work from implementation to higher-level activities like architecture, planning, and code review. Skills traditionally associated with staff engineers—maintaining context across systems, managing asynchronous workflows, steering AI agents effectively, and reviewing code—are becoming baseline requirements earlier in careers. Junior engineers now operate at senior-level abstraction while seniors architect at staff-level scale. Success depends less on prompting AI and more on managing personal context, maintaining focus during agent runtime, and developing new workflows that treat AI as a junior engineer to delegate to.

  6. 6
    Article
    Avatar of github_updatesGitHub Changelog·21w

    Copilot Memory early access for Pro and Pro+

    GitHub Copilot Memory is now in early access for Pro and Pro+ subscribers. This feature allows Copilot agents to learn from user feedback and actions, building repository-specific memory to improve assistance across coding and code review workflows. Users can enable it through Settings > Copilot, and GitHub plans to expand availability to more subscription tiers in the future.

  7. 7
    Article
    Avatar of cassidooCassidy's blog·23w

    The ABCD framework for feedback

    The ABCD framework provides a structured approach to giving and receiving feedback through four specific questions: what's Awesome, what's Boring, what's Confusing, and what Didn't you believe. This simple method helps elicit specific, actionable responses instead of vague reactions when reviewing blog posts, tutorials, projects, or products.

  8. 8
    Video
    Avatar of continuousdeliveryContinuous Delivery·23w

    We Tried Trunk-Based Development... The Results Were Shocking.

    A real-world case study from MFT Energy demonstrates how a team successfully adopted trunk-based development by committing directly to main with no branches. The team made ~2,500 commits (90% under 200 lines) while building a power trading bidding service under tight deadlines. Results showed main remained deployable (9/10 score), high quality (8.5/10), excellent DORA metrics, and strong team preference (8.5/10 would continue). Success came from small batches, fast feedback, and low transaction costs. However, non-blocking code reviews scored poorly (6/10, -7 NPS) due to lack of prioritization and unclear purpose. The team found trunk-based development worked well even without mandatory TDD or pair programming, proving that small increments with fast feedback create quality and stability.

  9. 9
    Article
    Avatar of mitsuhikoArmin Ronacher·21w

    A Year Of Vibes

    A reflection on how AI-powered coding tools fundamentally changed software engineering practices in 2025. The author transitioned from traditional coding to working with agentic tools like Claude Code, publishing 36 posts and having extensive conversations about AI. Key challenges emerged around version control systems not capturing AI-generated code context, code review workflows needing redesign for human-AI collaboration, and the difficulty of separating human intent from machine output. The piece explores the anthropomorphization of AI tools, the rise of terminal-based coding agents, and identifies opportunities for innovation in observability, version control, and establishing new social contracts for AI-assisted open source contributions.

  10. 10
    Article
    Avatar of zaidesantonManager.dev·24w

    The 'delayed opinions givers' - engineering teams everybody hates

    When another team needs work done in your domain, you face three choices: ignore them, give delayed attention, or prioritize helping them. Delayed attention is the worst option—it wastes engineer-weeks through blocked work and messy merges while gaining nothing. Teams that prioritize unblocking others (even at the cost of their own deadlines) build better reputations and create reciprocal relationships. The key is responding promptly to requests like code reviews and consultations, treating cross-team dependencies with the same urgency as internal work. Set boundaries with teams that abuse this approach, but default to being a 'giver' team that puts others' needs first 90% of the time.

See all Code Review archives