My New AI Co-Worker
June 27, 2025
Back to homeOver the past few months, I’ve fundamentally changed how I approach software development. What started as occasional experimentation with AI tools has evolved into something much more intentional—AI has become my daily coworker, and honestly, I can’t imagine going back.
The Shift from Passive to Active AI Usage
The key difference isn’t just that I’m using more AI tools—it’s that I’m intentionally seeking them out as solutions to specific problems. Instead of AI just “happening to me” through autocomplete suggestions or random ChatGPT queries, I’m strategically incorporating it into my workflows. Have I been more productive? I think so. More importantly, I’ve been able to tackle problems and achieve outcomes that would have taken me significantly longer—or might have been beyond my reach entirely.
My AI Toolkit
Cursor has been a game-changer. Cursor TAB is absolutely an amazing tool for code completion, but what really excites me are the emerging capabilities around MCPs (Model Context Protocols) that let AI interact directly with datasets, and Cursor Agents that can achieve large, complex outcomes quickly. I see these as force multipliers—tools that don’t just make me faster, but expand what’s possible within the constraints of a typical workday.
Where AI Actually Delivers Value
Inline Documentation for Learning
When I encounter code I don’t fully understand, I ask AI to add inline documentation explaining each piece. This helps me break down complex concepts quickly. Once I understand it, I have the AI remove the documentation. It’s like having a patient tutor who never gets tired of explaining things.
Deep Partner Debugging
This feels closest to actual pair programming. The AI walks me through different facets of a problem, and even when it doesn’t give me the exact answer, it helps me view bugs from perspectives I wouldn’t have considered. The multi-step debugging conversations can be surprisingly effective.
Solid Starting Points
AI-generated initial passes on markup or features can yield terrible results, but most of the time they give me something solid to build from—taking me 20-40% further than starting from scratch.
Large-Scale Refactoring
Recently, I needed to convert a project to use CSS logical properties. Instead of constantly looking up property mappings, I developed a workflow where I’d style features normally, then use AI to convert everything to logical properties afterward. This worked excellently and saved me from context-switching constantly.
Decoding Cryptic Errors
Error messages are often cryptic. Feeding them to AI—especially in an AI-enabled editor with file context—provides much clearer explanations of what’s actually going wrong.
Prompting Strategies That Actually Work
Through trial and error, I’ve developed some techniques that consistently improve results:
- Context is everything. Tell the AI what you’re working on. The more context, the better the answers.
- Be careful with thinking models. They can go down rabbit holes that have nothing to do with your actual problem.
- Include documentation. Even though these models have likely seen the docs, providing them as context helps the AI understand both what you want and how you want to achieve it.
- Keep requests focused. Small, compact requests work better than sprawling ones.
- Manage your chat context. Start new chats when changing context to avoid confusion. Return to old chats when revisiting previous problems—the context is valuable.
- Match the model to the task. I use smaller, faster models for straightforward changes and reserve the complex thinking models for larger scope requests. If a small model isn’t cutting it, I’ll escalate to a larger one within the same chat context.
- Screenshots provide valuable context for models with vision capabilities.
- Create rules for repeated prompts. If you find yourself typing the same thing over and over, consider making it a project rule or system prompt.
Staying Current Without the Hype
Model releases come fast and often. I try not to jump on every “new model bandwagon,” but I do hunt down models that are legitimately better than what I was using before. The key is being selective about upgrades that actually improve your workflow.
Looking Forward
We’re still in the early days of AI as a development tool, but I’m convinced we’re witnessing a fundamental shift in how software gets built. The tools that excite me most aren’t just faster versions of what we had before—they’re expanding what’s possible for individual developers to accomplish. My AI coworker doesn’t replace human collaboration, but it’s become an invaluable partner in the day-to-day work of building software. And honestly, that feels pretty remarkable.