@RayFernando1337: Opus 4.8 Max Thinking in Cursor with Multitask workflow is top tier at long context understanding, speed, and implement…

Opus 4.8 Max Thinking in Cursor with Multitask workflow is top tier at long context understanding, speed, and implementing large features in Swift.

Every subagent has its own context window and it really helps when working on complex workflows. Here are some of the ways I take advantage of their premier subagent harness:

1. When planning I tend to ask the agent to kick off subagents for comprehensive research. I do this for new libraries I’m not familiar with and it helps me learn how to integrate it into my app. I will also tell the agent to clone the repo and use it as a reference for future features.

1a. In addition, my two favorites for research tasks are Exa code (it’ll look at many more sources than the built in Exa search) and ref tools MCP.

2. I like to make the agent generate a comprehensive phased plan where I will do acceptance testing after each phase so we can review the results of the integration and make room for changes or logging issues with bugs that need to be addressed in later phases. (I’ll share my QA & Bug review skill in the thead post.)

3. Ask the agent to make a hand off prompt for a phase of your work based on your plan so it’ll give you a heads up on what it is thinking about implementing. That way you can ask questions or guide the agent’s work. This is key to learning on the fly with your agent and not letting your mind rot.

4. After testing by hand or doing a phase of work tell the agent that you’ll like to update the plan to keep track of the work needed and help you scope the issue to either be fixed in the current phase or roll it into a different phase of work.

I’ve tried a lot of workflows where I’m very hands off like /goal or missions from droid and I much prefer to be in the drivers seat to learn and drive my agentic engineering standards to a new high.

At the end of the day, as an engineer, you are responsible for what lands in the codebase and you’ll quickly feel out the vibe of your harness, model, and other factors.

Cursor’s latest agentic workflows are very well integrated with the models and I can’t wait to share how easy handing off workflows to the cloud are in the latest releases.

I’m currently kicking off a massive refactor to UIKit and will report back on how this goes.

The bugs that cause churn almost never show up in a diff, and you only really catch them when you stop reviewing code and start using the thing like a super user at runtime, which is about as close as you can get to your actual customer environment. That’s where I get a lot more value per token than a normal code review, and I’ve finally started writing my approach down.

I can’t wait to make my agents go BRRRRR. What hardware do you wanna see first??