GitHub Community
What AI Tools Are You Using in Your Daily Development Workflow? #206350
Discussion TypeProduct Feedback Discussion ContentHi everyone 👋 I’m curious about how developers are currently using AI tools in their everyday work. AI is becoming a bigger part of software development, from writing code and debugging to architecture design, documentation, testing, and automation. Some questions for discussion: Which AI coding assistants or tools do you use regularly? I’d love to hear everyone's experiences, opinions, and recommended tools. |
Replies: 4 comments 3 replies
|
💬 Your Product Feedback Has Been Submitted 🎉 Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users. Here's what you can expect moving forward ⏩
Where to look to see what's shipping 👀
What you can do in the meantime 💻
As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities. Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐ |
|
I mostly see AI as a productivity tool rather than a replacement for the development process. The areas where I find it most useful are debugging, explaining unfamiliar code, generating test cases, and handling repetitive tasks. It can also be helpful when I'm stuck on an approach and want to explore a few possible solutions quickly. I still prefer to handle important architecture and design decisions myself and carefully review generated code before using it. AI can produce something that looks correct while still making assumptions that don't fit the project. I think one of the most important AI-related skills in 2026 will be learning how to verify and review AI output effectively. Knowing the underlying concepts is still what lets you tell whether the generated solution is actually reliable. |
|
But, AI always make a lot of mistake. |
|
For me, AI has basically become a small development team living inside my editor 😂 I use different tools for different jobs:
But one thing that has probably improved my development speed more than any specific model: I talk to AI in my native language. 😄 I don't bother translating my thoughts into “professional developer English” before giving them to an AI. I explain the problem exactly how I would explain it to another developer sitting beside me. And honestly, it works surprisingly well. It makes communicating complex ideas much faster because I'm thinking about the problem, not about how nicely I can phrase the prompt. For complex development, I also rely heavily on low-level design documents. Before asking AI to implement something, I try to make the architecture, requirements, constraints, and expected behavior clear. I also use a kind of looping/fallback technique with Claude. I define success criteria, restrictions, and fallback conditions, then let it iterate and self-correct instead of expecting the first attempt to be perfect. I try to enforce good software design while doing this too. For example, I make use of patterns like Repository, Unit of Work, and Events where appropriate. The goal isn't just to get working code quickly, but to keep the codebase maintainable and reduce future issues. One thing I find particularly useful is maintaining a development history. After every significant prompt, the AI updates the documentation with:
So when a mistake happens, we don't just fix it and forget about it. The mistake gets documented so the same thing doesn't keep happening again. I still don't blindly trust AI with everything. I let it handle a lot of the repetitive implementation, exploration, and debugging, but I keep the final decisions around architecture, business logic, and critical code under my control. At this point, I don't really see AI as “replacing developers.” I see it more as removing the boring parts between having an idea and actually building it — while making sure we learn from every mistake along the way. |
|
Great insights. I think the low-level design document approach is one of the most important parts of effective AI-assisted development. Without clear architecture, constraints, and acceptance criteria, AI can produce code that works but creates long-term maintenance problems. I have found that AI works best when treated like a junior-to-mid engineer: give it context, define boundaries, review the output, and continuously improve the system knowledge. The idea of recording failed attempts and reasons behind decisions is especially valuable. It turns AI usage from simple code generation into an evolving engineering process. I’m curious if you have experimented with AI agents that can automatically maintain this kind of project memory across multiple tasks? |
|
The pattern in this thread — debugging, explaining unfamiliar code, generating tests — is exactly where I kept re-explaining the same standards to the agent every session: "review this properly, don't rubber-stamp", "find the root cause before patching", "write tests that actually catch bugs, not happy-path ones". What fixed it for me was moving those instructions into skills (the agent-skills format, Concrete example: a 7-dimension review skill (correctness, security, performance, contracts, error handling, tests, maintainability — findings ranked blocker / should-fix / nit) instead of the generic "looks good" a bare prompt gives you. Same discipline for debugging: reproduce → bisect → single hypothesis → minimal fix → regression test, with shotgun-patching explicitly banned. And test generation that targets boundaries and failure paths on purpose. I open-sourced the 7 most useful ones (MIT). They also install into Codex, Gemini CLI, OpenCode, Goose and other agents via the open skills installer: npx skills add Hahaknight/claude-skills-proRepo: https://github.com/Hahaknight/claude-skills-pro Curious what standards you catch yourself repeating every session — that's usually the best candidate to turn into a skill. |
|
This is a really interesting approach. I think the biggest limitation of AI coding assistants today is not their ability to generate code, but the lack of consistent engineering context and standards. Moving those expectations into reusable skills makes a lot of sense. It turns AI from a general-purpose assistant into something closer to a team member who already understands the team's workflow. I especially like the debugging workflow: reproduce → investigate → form a hypothesis → make a minimal fix → add regression tests. Too many AI interactions jump directly into patching without understanding the actual root cause. The idea of documenting mistakes and preventing repeated failures is also powerful. It feels similar to building an engineering culture around AI instead of just using AI as a code generator. I’m curious: how do you decide when a skill should be created? Do you create a new skill whenever you notice repeated corrections, or do you have a more formal process? |
I mostly see AI as a productivity tool rather than a replacement for the development process.
The areas where I find it most useful are debugging, explaining unfamiliar code, generating test cases, and handling repetitive tasks. It can also be helpful when I'm stuck on an approach and want to explore a few possible solutions quickly.
I still prefer to handle important architecture and design decisions myself and carefully review generated code before using it. AI can produce something that looks correct while still making assumptions that don't fit the project.
I think one of the most important AI-related skills in 2026 will be learning how to verify and review AI output effectively. Know…