Expediting AI / Claude Code Workflows | Sai Nimmagadda

Screen recordings as spec
Gemini
Use Gemini 3 pro for specs for example
Then follow up with Claude to interview you with the AskUserQuestionTool
“Read this @Spec.md and interview me in detail using the AskUserQuestionTool about literally anything: technical implementation, UI & UX, concerns, tradeoffs, etc. But make sure the questions are not obvious. Be very in-depth and continue interviewing me continually until it’s complete, then write the spec to the file.
A traditional agent with limited context often tends to write everything from scratch stemming from outdated context windows
Instead of using libraries, it tends to make them itself…
To resolve this -> Give the specification to something like chatGPT that does heavy thinking and prompt with searching enabled:
“Using this particular spec, can you search online and find any relevant GitHub packages that may be useful in developing out this feature? Search online and find well maintained packages that seem to be regularly updated, have reliable communities, and can essentially simplify the development of I this particular feature. When there are multiple options, propose different ones. They should be compatible with”
Have Claude Code break down the full spec into phases, and have it check off each phase in the spec
Workflow i.e Gemini 3 pro spec -> ChatGPT packages -> Claude Code Spec based implementation
Opus 4.5 for implementation most of the time.. Sonnet 4.5 for quick review & polish
GPT 5.2 for architecture and planning related decisions
When I know the spec is going to be big.. pass to GPT 5.2 to understand how to structure for example
Maybe try Gemini 3 Pro for Design and Creativity
Haiku 4.5 is fast … use it for quick answers for code, small helpers and explanations, targeted edits I know should be made
Look into Codex CLI
Multiple CI tools working on different tasks in different projects…
Git work queues merging can be a pain
When you enable planning mode, Claude code spins up explore sub-agents that explore for well established patterns, passes to planning sub agent that uses information, and creates a plan for future agents
Found frontend sub-agent, backend-agent, etc didn’t coordinate well, may misinterpret requirements, complex specifications to know when to use each sub-agent this way..
Within the same codebase, don’t do this maybe
Sub agents were better for controlling the context instead… ie
Can you spin up an opus sub agent, a YY sub agent that uses the X MCP server other than figure out
Consider this bug from multiple angles… etc

References:

https://www.youtube.com/watch?v=sy65ARFI9Bg&t=343s