When using AI as a specialist the goal is never to automate

Contrasting reactions to AI adoption at work.

Over the last two weeks I’ve began interviewing people about how they use generative AI at work, and I’m starting to see two opposing types of responses.

Generalists seem to love it, and specialists hate it.

Whatever you are good at, you will do better than the best language model. If you try to use AI to do something you’re good at, you will be disappointed. For example, language models write better than most people, but they’re much worse than a good writer.

The solution here is to use AI for the boring stuff.

Create examples, find blog post images, fill out forms, proofread, track calories, drafting emails, etc. It performs better than average on most tasks, and in a lot of situations, that’s good enough. You get to concentrate on what you do best without wasting time on the boring stuff.

The other approach is to use AI as a specialist.

With smart prompting and some good process design, you can use language models to improve tasks you are already skilled at. This approach takes patience. It involves slowly articulating your process and establishing clear frameworks for how you think about good work.

For example, if you meditate on your writing process and break it down into discrete phases, then…

• AI is great for capturing thoughts; you can record a long ramble-y idea as it occurs to you, and AI will clean it up for later.

• You can compile a collection of notes and have AI organize your thoughts into a coherent argument before you start.

• You can use it to triage your research by summarizing or interrogating long articles to see which ones are worth reading.

• It can help you explain a complex idea for a specific argument.

• AI is also great at tricking you into moving things forward when you get stuck. It can write a terrible next paragraph, which you then start to fix, and voila, you’re writing again.

• It’s also great at evaluating your final piece against your own criteria for good writing.

Everyone’s process will be different. By examining how you do what you do, you can begin using AI as a reasoning engine in interesting ways.

An important caveat here is that, when using AI as a specialist, the goal is never to get the machine to do the work for you. AI cannot give you good taste. You will never escape having to do the actual writing because the goal is not to fully automate, but instead to effectively augment.