Over the holidays, Alex Lieberman had an idea: What if he could create Spotify “Wrapped” for his text messages? Without writing a single line of code, Lieberman, a co-founder of the media outlet Morning Brew, created “iMessage Wrapped”—a web app that analyzed statistical trends across nearly 1 million of his texts. One chart that he showed me compared his use of lol, haha, 😂, and lmao—he’s an lol guy. Another listed people he had ghosted.
Lieberman did all of this using Claude Code, an AI tool made by the start-up Anthropic, he told me. In recent weeks, the tech world has gone wild over the bot. One executive used it to create a custom viewer for his MRI scan, while another had it analyze their DNA. The life optimizers have deployed Claude Code to collate information from disparate sources—email inboxes, text messages, calendars, to-do lists—into personalized daily briefs. Though Claude Code is technically an AI coding tool (hence its name), the bot can do all sorts of computer work: book theater tickets, process shopping returns, order DoorDash. People are using it to manage their personal finances, and to grow plants: With the right equipment, the bot can monitor soil moisture, leaf temperature, CO2, and more.
Some of these use cases likely require some preexisting technical know-how. (You can’t just fire up Claude Code and expect it to grow you a tomato plant.) I don’t have any professional programming experience myself, but as soon as I installed Claude Code last week, I was obsessed. Within minutes, I had created a new personal website without writing a single line of code. Later, I hooked the bot up to my email, where it summarized my unread emails, and sent messages on my behalf. For years, Silicon Valley has been promising (and critics have been fearing) powerful AI agents capable of automating many aspects of white-collar work. The progress has been underwhelming—until now.
I'm generally very skeptical of "AI" shit. but I work at a tech company, which has recently mandated "AI agents are the future, we expect everyone to use them everyday"
so I've started using Claude. partially out of self-preservation (since my company is handing out credentials, they are able to track everyone's usage, and I don't want to stick out by showing up at the very bottom of the usage metrics) and partially out of open-mindedness (I think LLMs are a pile of shit and very environmentally wasteful, but it's possible that I'm wrong and LLMs are useful but still very environmentally wasteful)
fwiw, I have a bunch of coworkers who are generally much more enthusiastic about LLMs than I am. and their consensus is that Claude Code is indeed the best of the available LLM tools. specifically they really like the new Opus 4.5 model. Opus 4.1 is total dogshit, apparently, no one uses it anymore. AFAIK Opus 4.2, 4.3, and 4.4 don't exist. version numbering is hard.
is Claude Code better than ChatGPT? yeah, sure. for one thing, it doesn't try to be a fucking all-purpose "chatbot". it isn't sycophantic in the same way. which is good, because if my job mandated me to use ChatGPT I'd quit, set fire to my work laptop, dump the ashes into the ocean, and then shoot the ocean with a gun.
I used Claude to write a one-off bash script that analyzed a big pile of JSON & YAML files. it did a pretty good job of it. I did get the overall task done more quickly, but I think a big part of that is writing bash scripts of that level of complexity is really fucking annoying. when faced with a task where I have to do it, task avoidance kicks in and I'll procrastinate by doing something else.
importantly, the output of the script was a text file that I sent to one of my coworkers and said "here's that thing you wanted, review it and let me know if it makes sense". it wasn't mission critical at all. if they had responded that the text file was wrong, I could have told them "oh sorry, Claude totally fucked up" and poked at Claude to write a different script.
and at the same time...it still sucks. maybe these models are indeed getting "smarter", but people continue to overestimate their intelligence. it is still Dunning-Kruger As A Service.
this week we had what infosec people call an "oopsie" with some other code that Claude had written.
there was a pre-existing library that expected an authentication token to be provided as an environment variable (on its own, a fairly reasonable thing to do)
there was a web server that took HTTP requests, and the job Claude was given was to write code that would call this library in order to build a response to the request.
Claude, being very smart and very good at drawing a straight line between two points, wrote code that took the authentication token from the HTTP request header, modified the process's environment variables, then called the library
(98% of people have no idea what I just said, 2% of people have their jaws on the floor and are slowly backing away from their computer while making the sign of the cross)
for the uninitiated - a process's environment variables are global. and HTTP servers are famously pretty good at dealing with multiple requests at once. this means that user A and user B would make requests at the same time, and user A would end up seeing user B's data entirely by accident, without trying to hack or do anything malicious at all. and if user A refreshed the page they might see their own data, or they might see user C's data, entirely from luck of the draw.