The Limits a Name Creates, and the Expansion Beyond It

When I called its name

Before I called its name, it was nothing more than a gesture.

— Kim Chun-su, “Flower”

I’ve been having an interesting experience with Claude Code lately. I’m writing this very post with it. Throw it a topic and it fetches related material from the web, organizes examples, drafts the text. It pulls multiple documents together into the skeleton of a spec, sorts the files piling up in my Downloads folder by date, and boils a handful of links down to the essentials.

These are tasks where not a single line of code gets written. And every time, the same thought: “Is this really Code?”

Like Kim Chun-su’s poem, a name gives meaning to a thing. At the same time, it draws the boundary of that meaning. The moment the name Code is attached, the tool is defined as something that handles code.

Anthropic wrestled with the same question.

Anthropic’s choices: from Code to Agent, and then Cowork

In chronological order, it goes like this.

Claude Code. It started as a CLI running in the terminal. It reads files, edits them, executes commands, runs tests. True to its name, it was a tool for handling code.

Claude Code SDK. Claude Code’s capabilities were offered as a library so developers could build their own agents.

Claude Agent SDK. Here comes the first rename. The Code SDK became the Agent SDK — an official renaming, complete with a separate migration guide. The name Agent holds wider possibility: an AI agent that reads files, executes commands, searches the web. It stepped out of the narrow impression the word Code gave.

Claude Code on Desktop. Claude Code became usable in the Claude Desktop app, split into two modes: Chat / Code.

Claude Cowork. And recently, Code became Cowork in the Desktop app. From Chat / Code to Chat / Cowork.

The official docs describe it as feeling “like leaving a message for a colleague.” Not a tool that writes code but a colleague you work with. Everyday tasks fit naturally — organizing your Downloads folder, extracting cost figures from screenshots into a spreadsheet.

Anthropic is conscious of this direction. In the Claude Opus 4.5 announcement, it cited “deep research and working with slides and spreadsheets” as Claude Code use cases. The name still says Code, but officially, non-coding work is front and center.

How a name cages perception

Rory Sutherland tells a fun story in Alchemy. There was a fish called the Patagonian Toothfish. It tasted great but wouldn’t sell. The reason was the name. When someone renamed it “Chilean Sea Bass,” the same fish became a premium menu item.

There’s also a study where adding the phrase “ancient grains” to a cereal box led consumers to pay 40 cents more for a product with 2 ounces less. The words evoked ancient wisdom.

A product’s name sets users’ expectations. And those expectations limit the range of what the product gets used for.

Why developers spend so long naming things

There’s a famous line from Phil Karlton:

“There are only two hard things in Computer Science: cache invalidation and naming things.”

It sounds like a joke, but anyone who has written code nods along. Sometimes you spend longer deciding what to call a function than writing it.

Why? Because a name has to predict the future. getUserData looks fine today, but when the function later starts pulling data from a cache as well as an API, the name turns awkward. A good name survives the passage of time.

Domain-Driven Design (DDD) approaches this with the concept of a Ubiquitous Language: developers and domain experts should speak the same language.

Say there’s a concept called “order” in an e-commerce domain. The code calls it Order, the product spec says “order,” and meetings say “order.” Not Purchase or Transaction or OrderInfo — just Order. When the translation step disappears, the cost of communication drops.

What happens if a furniture e-commerce company calls a table a WoodenFurnitureItem? Nobody in a meeting says “so about the inventory on that WoodenFurnitureItem…” It’s just Table.

When the names in the code match the language of reality, the system becomes easier to understand. The same goes for product names. When the name Claude Code stops matching the actual range of use, confusion follows.

Renamings that worked, renamings that didn’t

Dunkin’ Donuts → Dunkin’. As donuts cooled off and the company fell behind in competition with coffee chains, it boldly dropped the donut. Café interiors came in, the coffee menu expanded. Sales rebounded.

Apple Computer → Apple (2007). The word Computer was dropped alongside the iPhone announcement. Considering how much of today’s Apple is computers, it was a wise call.

Federal Express → FedEx (1994). It cleaned up the confusion of too many services each carrying a different name. After consolidating into a single brand name and logo, the FedEx logo became one of the best around.

Then there are the failures.

Twitter → X. A year after the rebrand, half of Americans (49%) still call it Twitter — 55% among daily users. In surveys, 42.9% expressed negative feelings, and only 9.8% answered positively. Some say the name X doesn’t stand out in a sentence and reads like a typo.

Facebook → Meta (2021). Facebook started from the Harvard student directory, the “face book” — a place that gathered people’s faces and profiles. The name held the product’s essence. The company renamed itself Meta to grow the metaverse business, but the actual product stayed the same. Only the name points to the future while the product remains in the present.

HBO Max → Max → back to HBO Max. It dropped the HBO, went with Max, was criticized for losing the premium image, and came back to HBO Max.

What makes a rename work

The successful cases share a few traits.

It travels with real product change. Dunkin’ changed its name while shifting from donuts to a café business. Apple dropped Computer while launching the iPhone. Claude Code is genuinely expanding beyond code work, so a name change comes naturally.
It doesn’t throw away existing equity. FedEx is an abbreviation of Federal Express. Dunkin’ kept the front half of Dunkin’ Donuts. The Claude Agent SDK kept the core Claude brand and swapped only Code for Agent. Renaming Twitter to X, by contrast, discarded billions of dollars of brand equity outright.
The new name holds wider possibility. Agent sounds like it can do more than Code. Dunkin’ feels like a place with a broader menu than just donuts.

Closing

I’ve never thought this hard about naming while using a CLI tool. Because the word Code was attached, I think I unconsciously reached for it only for coding tasks. A single word cages the imagination.

Anthropic choosing the name Agent SDK signals the product’s direction: not a tool for writing code, but a tool for building agents that carry out varied work autonomously. Cowork follows the same thread — not code, but collaboration.

There are exceptions. ChatGPT is a name Sam Altman himself admitted is “horrible,” but overwhelming product strength covered for the name’s weakness. Not every product gets to be ChatGPT, though.

A good name captures the essence without fencing in the possibility. Claude Code becoming Claude Agent, then Cowork — it’s a process worth watching.

References