Getting Dangerous with AI

Your ripped denim jeans scuff the floor while Pearl Jam plays on your Discman; you just arrived late to class. It's October 3rd, 1996. In your least favorite course, biology, you've been put in a team project with the smartest students in your class. You're going to fail the course if you don't get a good grade on this project. Your assignment: a presentation on mitochondria.

You go home and use this cool new technology to do your project: the World Wide Web.

In two hours, you're able to pull information from across the web. When you meet with your project team, you discover they each spent six hours at the library and found only a fraction of the information you provided. They decide to use your report as the backbone of the final submission.

How did you (a nearly failing student) spend less time and produce higher-quality work than some of the brightest minds in your class? You harnessed a new technology in a way that gave you a major advantage.

Fast forward to October 3rd, 2026, and the same thing is happening with AI. Today, most people use AI every once in a while via chat; only about 1 in 6 Americans use it every day.

To get the most out of AI, you need to understand and master a few key concepts. My goal is to bring casual AI users up to speed so they can get that same advantage at work.

If you have access to OpenAI (ChatGPT), Anthropic (Claude), Google (Gemini), or xAI (Grok), then you have access to the best. Ideally, you have enterprise access provisioned and managed by your employer. Tools built on top of these models, like Copilot, also exist, but it's generally best to go straight to the source for the most flexibility. Between these frontier labs, it doesn't matter much which you have; they're all amazing if used properly. People are generally the bottleneck, not the lab.

Next comes choosing a model. Most people never think about this, but it matters if you use AI every day. Choose the right model for the job. You don't need a flamethrower to light a candle; you want a lighter.

For most tasks, use a medium-effort, cost-effective model like Sonnet for Claude or Terra for ChatGPT. This is your volume player; you can do a ton of work on it, and it won't burn through your usage too fast. For a highly complex task where you need a brainy counterpart, go up a level to Opus (or Fable for the hardest work) for Claude and Sol for ChatGPT.

Context is probably the most important and least appreciated concept for getting the most out of AI. Think of context exactly as the word implies: the background on a problem, its constraints, and possible solutions. The best practice is to think of context more intentionally as something you build and pass to AI.

Here are some context repositories that may exist today:

So does that mean you can just give AI access to everything and it will have all the context it needs? Not really. While full access is better than nothing, curating information for AI is even better because AI struggles with conflicting approaches and ideas. Imagine giving AI access to a Google Drive with 100 financial models, some rough drafts, others final versions presented to the board. AI won't know which is which and will pull them all into context.

Even better is having a folder called "Final Versions" with only the vetted models you presented. Point AI at that folder instead of the whole drive.

Best in class is making these folders or context repositories across everything you do, then pointing AI at them. This is a never-ending battle; you'll never finish managing context, you'll just master organizing it.

Most people use the chat feature of these models and call it a day, often copying and pasting into other systems. Instead, you can connect AI to those systems and have it do that for you. 

Most business applications now have built-in ways to let these models connect (Connectors) and interact with them through a standard called MCP (Model Context Protocol).

I recently rebuilt our product newsletter, which I used to copy over to our email marketing tool. I set up a connector and can now have AI build the entire newsletter from scratch; then I just edit it in the marketing tool. Not only is it faster, it's also better quality.

Model makers offer tons of out-of-the-box connectors that take less than a minute to set up; common ones include Gmail, Google Drive, Salesforce, GitHub, PowerPoint, and basically any business app you use. These are controlled by your organization's admin, so it's worth knowing what is connected today and requesting tools that would make sense for you to connect.

If you've ever been in charge of creating a PowerPoint deck on a recurring basis, you understand the fundamentals of a skill. Over weeks and months, through feedback from others, you make a better and better deck. As you make mistakes, you discover the unspoken rules, what works, and how to do it faster. Skills in AI work the same way: they're extra context on accomplishing a specific task.

If I were tasked with a monthly board deck, I would first create a board deck skill, and every time I worked on the deck, I would update the skill with what I learned. AI can now update skills for you, but it's often on you to decide when a skill is needed. Anytime you're doing something repetitive, make a skill!

If you master skills, you master the recurring task, which means your job becomes less routine and more about higher-level thinking. Most people will not be able to make this transition, but the ones that do will be immensely valuable.

Once you have model selection, context, connectors, and skills mastered, your productivity will reach new heights. But it's still tied to you being at your computer. 

Cloud-based agents and routines change that: you get a virtual computer in the cloud with proper access that can run tasks on a schedule. That means something like checking your email every hour and replying to requests can now be automated.

The real skill isn't automating everything; it's identifying what should not be automated. In a world with so much AI, the human touch in things like sales and relationship building will be more important than ever. That is the future of work for humans.

Thirty years ago, a new tool turned a failing student into the backbone of the team. This time the tool is AI, and the advantage goes to whoever learns it first.