How much time does this demand?
Introduction
A deep dive into understanding how much reliance to put on AI based on the goals of the task for learning and building.
"Oak's words echoed… There's a time and place for everything, but not now."
Like many engineers today, I've been using AI extensively in my daily work. In just the past week, I've used Cursor, ChatGPT, Claude, Perplexity, Warp, v0, Notion AI, Granola, and Dia. These tools have drastically changed how I access information and approach building products. I can now research topics by having Perplexity summarise 10 articles for me, and Dia can distill an hour-long YouTube video into a one-minute read. Projects that would typically take weeks or months, such as completing Advent of Code in Ruby or building an app for finding favourite beers, can now be done in days or hours. The main problem? I've retained very little information. I wouldn't even know how to create a "hello world" app in Ruby, and the pouring.at codebase is now a mess that cursor can't navigate anymore.
Why Not Everything Deserves the Same Attention
Not every task demands the same depth of attention or time investment, especially in a world where AI tools are always within reach. The first step in deciding how much energy to devote to something is getting clear on your underlying goal. Are you aiming to truly master a subject, quickly solve a problem, or simply check a box and move on?
Understanding Your Goal: Mastery, Efficiency, or Just Enough?
If your goal is to build deep expertise, whether learning a new programming language, writing a thoughtful essay, or developing a creative project, merely scratching the surface or depending too heavily on AI might prevent you from truly understanding the topic. Studies show that hands-on struggle, reflection, and even frustration are often where real learning happens. In these cases, it makes sense to slow down, ask questions, and resist the urge to outsource too much to automation.
On the other hand, if you're trying to maximise efficiency for repetitive, time-sensitive, or low-stakes tasks like summarising meeting notes, generating quick outlines, or handling administrative work, AI can be a powerful accelerator. Here, the goal isn't deep learning or personal growth, but freeing up mental bandwidth for what matters more.
There's also a middle ground: sometimes you need just enough understanding to make an informed decision or participate in a conversation. Skimming an article, listening to a podcast, or asking an AI for a summary can be the right approach. The key is matching your method to your desired outcome, rather than defaulting to either deep work or instant answers.
Ultimately, being intentional about your goals helps you decide when to dig in, when to delegate, and when to let go, so you can invest your time and attention where they'll have the greatest impact. My friend Melis recently wrote a post about serious reading, how showing up for a book and giving it your full attention is a real commitment that goes against our usual habit of skimming and moving on. That kind of focus isn't just for reading; it's a valuable approach for anything you want to do well.
The Role of AI: When to Use It (and When Not To)
AI tools can significantly reduce cognitive load, but they may also hinder the development of critical thinking skills. In my experience, the more I use Cursor, the more I find myself disliking it—not because it's a bad tool, but because it can remove much of the fun from the process, especially when it makes errors and you end up doing things manually anyway. That said, it's not always detrimental and has saved me considerable time in certain areas by reducing cognitive load. The best uses I've found are:
- Working on boilerplate-heavy projects that require multiple files to be created in specific ways, such as adding a new endpoint to a backend service—create a Cursor rule that references all relevant files with steps for adding these new files.
- Writing tests. I've never found much joy in writing tests, so delegating this task is a huge win in my book.
- Adding content. I recently created an app to visualize the value of credit card rewards for Yonder. Since they don't have a public API or share this information outside their app, I had to manually input all the data. Fortunately, with Screen Mirror and a Cursor rule, I could simply provide screenshots of the data instead of typing everything manually.
- Prototyping. Tools like Cursor and v0 are perfect for prototyping when you want to quickly test and verify interactions without worrying too much about appearance, code quality, or fully understanding the implementation—you can get a working feature out for testing very quickly.
In the recent yonder-experiences project, I used AI for brainstorming data visualisation ideas and gathering data, but then reached for pen and paper, sketched out my ideas, and built the app from scratch without agent mode or prompting, though I still made good use of auto-complete and cmd+k, which preserves more intentionality. I was much happier with the result compared to projects I've "vibe coded" in the past, and I understood the codebase far better.
The Value of Presence
"We want to keep the human in the loop. We just want to automate all that busy work… you should really have that final control at the end of the day, and we should design tools so that humans stay in control otherwise I think we're in trouble." — Chris Pedregal at AI Demo Days
I'm a big fan of Chris's outlook on keeping humans in control of AI tools. Many tasks we perform with AI are similar to delegating work to a colleague who completes it based on their interpretation. This works for some use cases but shouldn't be the default approach. When using AI, remember to be intentional about aligning your usage with your goals.
Practical Takeaways
Before starting a new project or task, pause and ask yourself: What's my real goal here? Do I need to develop deep understanding, or is skimming sufficient? Would using AI help or hinder learning and quality? Experiment with your workflow to find the combination of tools and approaches that leaves you feeling both productive and satisfied with the outcome.
I'd love to hear your thoughts and stories on this topic. Feel free to reach out on X - @dbillson