I've been thinking a lot about how to integrate AI into my daily workflows — which tools, which models, when, how much, to automate or to augment.
Admittedly, working in tech and being inundated by AI billboards all over San Francisco accelerates such inquiry. But I'm also convinced that every one of us will grapple with this soon.
The pace of change and overflowing marketing budgets don't help. Every time I felt like I'd finally "mastered AI," a random post would promise that "everything just changed" or "XYZ was just killed by ChatGPT" (hint — it wasn't). And I'd be off to the races, frantically downloading new software, hammering away at prompts — privacy and frugality be damned.
After several rounds of this, I've also come to realize that while "true tool homeostasis" is a while away, adopting evergreen principles helps me approach every new "revolutionary breakthrough" with more curiosity than adrenaline.
The Principles
1. Concepts are few, applications are many. There's 100% a Pareto curve here. LLMs are the core. Agents are just LLMs with enhanced capabilities, most of which can be summarized as remember prior conversations, use external tools, and operate in multiple steps.
2. Treat upskilling with urgency, but not as an emergency. Adrenaline is not your friend. Yes, these are must-know concepts. But so is Excel. And Powerpoint. AI isn't going anywhere and isn't shape-shifting overnight — 99% of it is still chatting with a bot. Improvements are incremental, iterative, and blown out of proportion.
3. Identify the task of least resistance. Change is hard. If you don't know where to start, pick one tiny workflow and implement it. Motion begets motion and progress compounds.
4. Remember that productivity is subjective. I no longer feel despondent if I don't produce ten thousand digital widgets a day. Productivity for me might mean quality over quantity, and that's okay.
5. The learning curve is far less steep than you think. I can't overstate this. Yes, terminals and coding agents may seem uncomfortable if you've never written software. But these are incredibly intuitive products designed by some of the best product minds in the world. They want to make them easy and universal — Claude Cowork is a great example.
6. What you don't read is as important as what you do. Occam's razor is your best friend. More on this soon.
7. Just start. Treat as axiomatic that there is no better way to learn AI than to use AI. Start where you are. If you don't use ChatGPT, then download it today. If you're building n8n workflows, try your hand at a React agent in Python. High-quality courses are a great starting point as well.