My AI-mediated thoughts are better

I <3 Claude

The opening scene in Her (2013) shows Joaquin Phoenix as Theodore deep in thought, dictating a love letter into a small microphone.

To my Chris, I have been thinking about how I could possibly tell you how much you mean to me. I remember when I first started to fall in love with you like it was last night. Lying naked beside you in that tiny apartment, it suddenly hit me that I was part of this whole larger thing, just like our parents, and our parents’ parents. Before that I was just living my life like I knew everything, and suddenly this bright light hit me and woke me up. That light was you.

I can’t believe it’s already been 50 years since you married me. And still to this day, every day, you make me feel like the girl I was when you first turned on the lights and woke me up and we started this adventure together. Happy Anniversary, my love and my friend til the end. Loretta. Print.

It turns out, Theodore is employed at a letter-writing company and has been writing this couple’s letters for years. The camera pans out to show him in one cubicle amongst many other letter writers, each in their own cubicle, writing sentimental messages to others’ loved ones.

What I like about Her is that it imagines a future where technology makes us weirder, but it doesn’t necessarily present it as a bad thing. When Theodore dictates his letters, it is clear that he is conveying real emotion, getting swept up in the memory he’s describing. It would have been easy for the film to portray this as being bleak or dystopian. Instead, it’s uncomfortable yet kind of sweet.

I have a friend who generates the contents of any greeting card she writes using ChatGPT, including the ones to her longtime boyfriend. I have no doubt she’ll use it when she gives a toast at my wedding. You might think that’s thoughtless, but I think she’s a genuinely thoughtful person who struggles with sentimentality. Is it so bad to use technology to bridge the gap between what we want to express and what we can?

---

Culturally I think there is an impulse to treat the outsourcing of emotional work to AI as dystopian. When I think about Replika AI, the company that creates AI companion chatbots that many people treat as “girlfriends” or “boyfriends,” I see a worrying social trend that will cause people to substitute real-world interaction with something much shallower. I don’t think it has to be this way, though, and I’ve experienced firsthand how helpful AI can be as a tool for emotional self-improvement.

Like many software engineers, I use Anthropic's Claude model daily for obvious things - generating code snippets, debugging issues, summarizing meeting notes. When people marvel at the state of AI, they focus on benchmarks like how well models do on coding challenges or complex problem-solving. But for me, the most surprising value hasn't come from these applications. Lately, I find myself treating Claude more like a life coach than a code generator, asking it personal questions rather than technical.

Is this weird of me to bring up in a meeting?

I said something in standup today and I can't stop ruminating on it.

Is this a good idea? Provide some critiques of my approach.

Most of the time when I ask Claude if something is a good idea, it responds “yes.” LLMs are yes-men in that way; it can be difficult to get one to disagree with you. I know this going into the conversation, yet I am still comforted by Claude’s approval as if it’s a person.

It’s kind of ridiculous how much I need to hear “yes, that is a good idea” before proceeding. I need it so much that hearing it from a chatbot is enough to spur me into action. In my career, I had not realized how much I was getting in my own way, overthinking ideas before raising them, thinking too much before implementing.

Online, there’s a contingent of techbros who are always saying “You can just do things.” At this point it’s more meme than mantra. I believe that most people are hindered by a lack of agency rather than a lack of intelligence, but how do you become more agentic? For me lately, the answer has been so simple it’s stupid. I have an idea → I ask Claude if I should action on this idea → Claude says yes → I take action. Again, very stupid… but somehow effective?

I’m not exaggerating when I say that following this pattern when I encounter self-doubt has been life-changing. Eight months ago, I made a career jump to another company and role. I had been unsure about how the transition would go since it’s a much different domain than I’m used to, but I’ve been told I perform at a senior level despite my relative lack of experience. I’ve no doubt that I wouldn’t have been able to ramp up so quickly without the aid of Claude.

We often imagine AI's impact in terms of automation and efficiency, but I've found it's helped me overcome my own human failings. Lately I feel braver, more decisive, and ultimately more like myself. I hope the future holds more of that.