Agent obsession is slowing people down. My clients in law, design, operations, and the arts who are eager to get good at AI imagine agents as capable go-getters inside their computer making work happen pristinely on their behalf. They talk of agents as a magical reality just out of reach. I’ve talked to numerous mid- and senior-level leaders fixated on what agents could do for them while missing the tools they can use now. I emphatically recommend you stop obsessing over agents and write skills you can use today.
Agents are like Clouds. We use them all the time already.
“Agent” names something real but works as a metaphor. Think about “the cloud.” How would you define “the cloud”? You might know that you use the cloud all the time, but stumble when asked to define it succinctly on the spot. The cloud seems like it must be a positive, useful thing… Just like agents! Feels similarly vague and useful, right?
“The cloud” is a metaphor for computers in different places storing data and software that your computer can use. “Agent” is a metaphor for an AI that can use tools and information to get things done without having all the intermediate steps spelled out. Just like you use the cloud constantly without thinking about it, you are using agents all the time if you are using workaday AI products. So, agents are already part of your life. Check, done.
I have watched half a dozen people who speak breathlessly about driving AI adoption at their organizations go blank when asked, kindly in private, to define what an agent is. That’s a good enough sample size for me to anecdotally conclude that something weird is going on. I encourage everyone to take a breath and put the hype words down.
What should you focus on then if it’s not agents? Skills! Skills are sets of instructions that you can define and use over and over again with various AI systems. They are really handy for a few reasons. Most of the workaday systems, like ChatGPT and Claude, understand how to use them. They are super flexible and written in plain human language. It can feel daunting to start because the possibilities are very much wide open. Making and using skills is deceptively simple compared to the promise in agent hype. It feels like you must be missing something the first time you make one. Do not second-guess yourself. Just keep iterating and experimenting.
Four of My Skills
Below I’m going to share four of my skills in detail to serve as inspiration. I’ve had C-suite clients and compliance officers literally swoon watching my skills in action. And then coo themselves when they got an example skill I handed them working on their own computer. Friends with computer science degrees want to see the details of how my skills work. This is good stuff, and you can do it, too. I myself have learned best how to use AI by learning about other people’s setups through blog posts and podcasts. I’d like to reciprocate here, to pull all you wonderful people out of the fantastical agent future and into the joyful skill-pilled present.
The technical details of how your skills work will likely be different from mine. That’s fine. The point here is to show you specifics so you get inspired about what skills will make sense for you.
People Say
This skill is the most self-contained and obviously useful. I simply ask the AI what people say about something, and it searches and summarizes relevant organic discussion on the topic, ignoring all other sources, especially marketing. Recent uses of this include: “What do people say about the best portable speakers for taking to the beach?” “What do people say about how to eat lunch in midtown Manhattan without hemorrhaging money?” “What do people say about AirTrain reliability and disruptions at JFK right now?”
Write Better
LLMs write in a very annoying way. It frustrates the hell out of most of us. This skill does its best to make the AI write better. Even with it in place, I often complain back at the AI for sounding ridiculous. The AI logs my complaints, which we review together to revise the skill periodically. A lot of the technical complexity came from giving feedback on when it succeeded and when it failed. I didn’t even realize how complex this skill got until I sat down to document it for you all.
Work Buddy
This skill was something the AI came up with for me. I was skeptical at first, but it’s working really nicely. I’m thrilled. The skill puts work blocks on my calendar for various annoying and stuck tasks that just need a bit of focus. At the scheduled time, it helps me actually get the thing done by organizing information needed to get going, presenting it to me in chunks, and teeing up the immediate first thing to do. It also keeps track of its own success rate in helping me get things done, for consideration in improving itself over time.
Deep Reckoning
This skill came about because I worked through a really stuck life problem over a number of involved sessions with the AI’s help. The results were so illuminating that I wanted to be able to repeat the process on other topics in the future, so we made a skill. This one is extremely laborious for me to run. It’s only appropriate for chronically stuck, challenging situations. The skill’s first step is to size up whether the issue is even a good candidate for the skill. If it is, the skill clarifies the core issue, helps me assemble relevant personal materials and helpful outside content from reputable sources related to the challenge, reasons over all that to offer a range of recommendations, gets my feedback on those recommendations, revises accordingly, plays the recommendations forward, and assembles the whole thing into a beautifully designed webpage for me to review and reference. Like I said, this one is involved.
Things to Notice about these Skills
Some of them I thought up (people say), some of them the AI did (work buddy). Some were a product of an annoyance I wanted to address (write better) and some came from something working well that I wanted to make repeatable (deep reckoning). Some are single files of plain English (work buddy). Some are whole folders with various notes and custom scripts (write better, people say). They all are bespoke to my needs and my system. They all go through constant iteration and refinement.
How You Can Get Started
In most AI systems, you just tell the chatbot you want to make a skill and it will help you out. You could point the AI at the official standard documentation on skills to help it out or even link to this essay or one of my skills as reference to get it started. That’s it. Once the skill is made, you just trigger it, often in plain English, while you talk to the chatbot. When you want something about the skill changed, you tell the chatbot what needs to be improved and it makes the edits for you. Yes, really, that’s it.
Good luck, have fun. May you make skills and prosper. The agents are in there working for you too. I promise.