Kieran Klaassen

AI-assisted code review using parallel agent review is more effective than manual line-by-line code reading, catching critical bugs while reducing review time from 40 minutes per 200 lines to 15 minutes for complex 1,000+ line changes.

“Author of "I Stopped Reading Code. My Code Reviews Got Better." in Source Code”

Compounding engineering—building self-improving development systems where each code review and bug fix teaches the AI to improve iteratively—enables AI tools to learn team preferences and prevent entire categories of bugs over time.

“Author of "My AI Had Already Fixed The Code Before I Saw It" in Every”

Planning with AI rather than rapid coding produces better features and teaches AI systems to understand developer preferences for future requests.

“Author of "Stop Coding and Start Planning" in Every”

Using AI agents to plan before coding—by running parallel research operations—helps engineers avoid building wrong features and ship faster than jumping directly to implementation.

“Author of "Teach Your AI to Think Like a Senior Engineer" in Source Code”

Claude Haiku 4.5 delivers nearly Sonnet 4.5-level performance at 3x cheaper pricing, making it an optimal choice for developers building agentic applications.

“Co-author of the article; tested Claude Haiku's math capabilities with Uber bill calculations”

Droid, Factory's agentic coding product, enables developers and non-coders to switch between GPT and Claude mid-task to optimize for each phase of work without switching tools.

“General manager of Every's AI email assistant Cora who initially dismissed Droid but converted after discovering its multi-model capabilities and subagent features.”

Claude Opus 4.1 outperforms competing models like OpenAI's o3 and Google's Gemini 2.5 Pro on specific tasks, particularly excelling at autonomous coding, honest editing, and long-form task completion without intervention.

“Cora general manager and Rails expert providing feedback on Claude Opus 4.1 coding capabilities”

OpenAI's DevDay 2025 launched AppsSDK and operator features, but lacked groundbreaking announcements compared to prior years, suggesting the company is optimizing existing opportunities rather than pushing innovation frontiers.

“Observed that AgentKit is a workflow builder rather than an agent builder for business users”

“Kieran Klaassen stress-tested Opus 4.5 by running 11 projects in six hours.”

“Team member who tested Opus 5.5 for coding and design work, providing performance benchmarks and assessments.”

“Testing engineer at Every who ran LFG coding benchmarks on frontier models.”

“Manages OpenClaw agent named Klont that adapts to owner's preferences like recommending breathing exercises.”

“Kieran Klaassen created the compound engineering methodology and discusses human roles in AI workflows.”

Anthropic's Sonnet 4.6 delivers Opus-level performance at half the cost ($3/$15 per million tokens), making it viable for production applications previously too expensive to run on Opus.

“Tested Sonnet 4.6 model and teaches compound engineering philosophy for AI-native product development.”

Anthropic launched Claude Cowork, a visual interface bringing Claude Code's agentic workflow to non-technical users, built in one and a half weeks and available exclusively to Claude Max subscribers on macOS.

“Co-led testing and livestream demonstration of Claude Cowork.”

Anthropic released Claude 3.7 Sonnet, a hybrid reasoning model with dual modes of thinking, and Claude Code, an agentic coding tool that is particularly powerful for development tasks despite not yet being fully production-ready.

“Provided feedback that 3.7 is great but not ready for production work, while Claude Code is impressive for new projects.”

Claude Code enables a solo developer to ship features at the velocity of a five-person engineering team by automating code writing while the developer focuses on architecture and product thinking.

“Demonstrated how Claude Code enables him to ship code faster than ever by having AI write 100% of his code in the last two months.”

GPT-5.2 is an incremental upgrade that excels at extended knowledge work tasks (70.9% on GDPVal vs 38.8% for GPT-5.1) but offers minimal improvement for everyday chat, with real gains expected when powering agentic tools.

“Tested GPT-5.2's compatibility with Cora and found it excels at following instructions but is less resourceful than other models.”

Anthropic's Opus 4.5 represents a world-changing shift in AI-powered coding, enabling developers to build complete applications without hitting error walls that plagued previous models.

“Called Opus 4.5 the most exciting model maybe ever and started 10 projects at once using it.”

Anthropic's Opus 4.5 enables autonomous app development without writing code, collapsing months of developer work into days through agent-native capabilities and advanced reasoning.

“Co-hosting a Claude Code Camp on Opus 4.5 for paid subscribers.”

Anthropic's Claude Opus 4.5 is the best coding model available, capable of handling extended 'vibe coding' tasks without reaching performance limits that plague competing models from Google and OpenAI.

“Tested Opus 4.5 extensively on coding and found it significantly outperforms Gemini 3 and Sonnet 4.5 in coding tasks.”

Anthropic's Claude agents feature enables AI systems to create other agents independently, representing a shift toward 'compounding engineering' where systems build systems iteratively.

“Test-drove Claude's newest agent features by telling agents what to do using AI voice tool Monologue.”

Factory's Droid coding agent enables seamless switching between different AI models (Claude, GPT) without context loss, making it superior to individual subscription plans for production engineering workflows.

“Part of the team walking attendees through how to use Factory AI's Droid tool at the Droid Camp event.”

Anthropic's new cloud and mobile version of Claude Code shows promise but is not yet ready for daily use, lacking features like inline code diffs and seamless task transition between devices.

“Uses Claude Code running multiple instances in parallel as part of his compounding engineering workflow.”

Claude Sonnet 4.5 is 50% faster and more steerable than previous Claude versions, excelling as a day-to-day coding tool, though GPT-5 Codex still outperforms it on difficult production bugs.

“Tested Claude Sonnet 4.5 and found it about 50 percent faster than previous versions, and it solved a bug in 20 minutes that Opus 4.1 couldn't crack.”

OpenAI's Codex is a coding agent designed for senior engineers working on existing codebases, enabling autonomous feature building and bug fixes with clean code output, though its utility is limited for new product development.

“General manager of Cora who built the product's codebase with over 5,500 commits and tested OpenAI's Codex coding agent.”

DSPy is an automated prompt-optimization framework that has made traditional manual prompt engineering largely obsolete, with the author stopping manual prompt writing in favor of DSPy-optimized prompts.

“Participated in learning DSPy prompt optimization framework alongside other Every team members.”