The Great AI Coding Tool Debate An Ask HN post from June 20th sparked a revealing conversation about which AI coding assistants developers are actually using in their day-to-day work—and the answer is more fractured than you might expect. The question was simple: "Do you use Claude Code, Codex, or a different vibe coding/agentic engineering tool for most of your work? Why?" What followed was a window into how heterogeneous the AI tooling landscape has become.
Claude Code vs Codex: The Preference Divide The comments reveal clear tribal lines. One developer noted their organization used to pay for both but preferred Codex around three months ago—though they acknowledge finding "Claude significantly lazier or requiring significantly more guidance." That said, they conceded Claude tends to have better design and UI/UX intuition. Another commenter summed up their choice simply: "claude - org pays for it," which tells you everything about how enterprise licensing drives adoption.
The Hybrid Approach Is Gaining Traction Perhaps the most interesting pattern emerging from this thread is developers running multiple AI agents simultaneously for different tasks. One developer described using GPT 5.5 High through Codex for brainstorming and solution design, then switching to Claude Opus 4.7/4.8 for implementation work. They reported being "quite productive" on greenfield rewrites and refactors, though acknowledged bug fixing remains "more hit and miss." Another commenter uses OMP (Open Memory Protocol) with subagents—deploying different model configurations like ds4pro/mimo for easy features versus 5.5/4.6 for larger, harder problems.
The Security Factor in Enterprise Adoption When asked why they stick with Claude Code and Codex rather than open-source harnesses like OpenCode that support those same models, one developer offered a candid response: "Security is a thing here now so sec guys feel more comfortable with big names rather than 'random' open source." They also noted subscription costs favor established players—"subscriptions take you a long way for not very much." This suggests enterprise adoption of AI coding tools isn't just about capability but liability and vendor accountability.
The CLI-First Crowd Not everyone is buying into purpose-built agents. One developer mentioned having access to Claude and Kiro at work but uses everything via CLI because they find it "the most flexible setup"—CLAUDE can search anywhere, read files, write code, and execute programs without agentic scaffolding getting in the way. This echoes a broader philosophy among veteran developers: tools should augment existing workflows rather than replace them.
The Pen-and-Paper Contrarian The thread's most provocative comment came from someone claiming they use "my brain, pen and paper" because they work on "serious projects that are of high risk and are mission critical." They added a warning: "There will be a time where companies like Anthropic will disallow you from using AI in their interviews. So you are cooked if you are over-reliant on coding agents." Whether that's prescient or just gatekeeping, it captures the tension between efficiency-seekers and those worried about skill atrophy.
Key Takeaways
- Claude Code and Codex remain the dominant enterprise choices, but preference splits roughly evenly based on laziness vs. design intuition
- Multi-agent workflows are emerging as a best practice—using different models for brainstorming versus implementation
- Enterprise security concerns favor established vendors over open-source harnesses despite feature parity
- CLI access remains preferred by power users who want flexibility over abstraction