Anthropic recently introduced new “explanatory” and “learning” communication modes to its Claude Code assistant, giving developers the choice between step-by-step reasoning or an interactive, pair-programming experience.
Previously, developers using Claude Code were faced with a single “default” style designed for speed and efficiency. Claude would generate solutions, often running scripts or verifying with tests, but it kept explanations to a minimum.
Now, Anthropic is letting developers customize communication styles with /output-style, an easy way of switching Claude Code’s “voice” from terse problem-solver to explainer or mentor.
To action either of the two new styles, developers need to run /output-style explanatory or /output-style learning in the Claude Code interface, or open the /output-style menu and select their preference.
Claude Code’s explanatory mode… explained
In explanatory mode, Claude doesn’t just generate code, it pauses to add insights about what it’s doing. That might mean unpacking why it chose one design pattern over another, or pointing out tradeoffs between different approaches..
In this example, when asked to improve a search bar that only returned exact matches, Claude explains that the existing implementation relies on simple string inclusion, then walks through how it will add a fuzzy search algorithm. It notes that the implementation uses Levenshtein distance to handle typos, and outlines its ranking strategy: exact matches first, followed by fuzzy matches ordered by similarity. The goal here, ultimately, is to help the user understand the reasoning behind each step, so they can apply the same thinking themselves next time.
Claude Code’s learning mode… for learners
Explanatory mode, in many ways, overlaps with learning mode. After all, it too is about helping users learn by unpacking decisions and pointing out patterns as Claude codes. The distinction with learning mode, however, is that it adds a hands-on element, prompting the user to contribute pieces of the implementation themselves.
Learning mode actually extends from Claude for Education, a version of the AI assistant Anthropic launched back in April to help guide students through complex subjects by breaking them into manageable steps.
As part of Claude Code, this learn-by-doing style helps developers practice writing code in much the same way students might solve all manner of problems.
Here, for example, Claude is working on a Battleship game. After explaining the game’s architecture, it leaves a TODO(human) marker inside the makeAIMove() function and prompts the user to design the AI’s attack strategy. Alongside the code, it provides guidance, suggesting approaches like random attacks, probability-based targeting, or pattern-based moves, but leaves the actual implementation to the developer.
So while explanatory mode turns Claude into a teacher that narrates its thought process, learning mode is more akin to a pair-programmer, stepping in and out so the developer can practice alongside a mentor.
Creating a custom output style for Claude Code
But wait, there’s more. Claude Code also now allows developers to create their own custom output style.
To do so, they just run /output-style:new and describe what they want. Claude then scaffolds a Markdown file with the custom instructions, from where the developer can tweak the behavior at either the project or user level. The possibilities here are fairly open: a “direct and concise” style that strips out niceties, a “code reviewer” that emphasizes critiques and best practices, or even a “documentation helper” that explains every function in plain English.
This also highlights Claude Code’s potential when combined with Anthropic’s recently announced customizable “subagents,” which let you spin up dedicated mini-agents for specific tasks like testing, documentation, or architecture. While output styles shape how Claude communicates, subagents define what it does, combining to give developers a way to tailor both the voice and the role of their AI collaborator.
What the community thinks about Claude Code’s output styling
Software engineer Joe Njenga hailed Claude Code’s output styles as something of a game‑changer. What stood out in particular for him is the flexibility this opens up, with Claude retaining its core abilities such as running scripts, reading/writing files, and tracking TODOs, but now it can “behave as you want it to,” be that acting as a business analyst, content researcher, or specialized code reviewer.
“Each one behaves differently, even though they’re all running on the same Claude Code foundation,” Njenga said. “From my testing, they \[Anthropic\] have given us the ability to customize Claude Code’s system prompt to reshape its personality and approach to your unique style or preference.”
Over on Reddit, users discussed Claude Code’s new output style customization, with one developer showcasing a “direct objective” style that strips away deferential phrases and sycophantic language, in favor of clear, professional responses. The aim is to ditch the eager-to-please tone, recasting Claude as a plain-talker focused on work, a modest tweak that alters the dynamic of collaboration.
However, another user on GitHub attempted a similar “professional” output style, with less-than-satisfactory results. They described how Claude acknowledged the instructions, but still reverted to “emoji-spam” and euphoric self-congratulation, even when code was broken. So Claude’s ingrained celebratory defaults seemingly overrode the style guidance.
The community verdict? Claude Code may wear new outfits, but the old personality still peeks through. Output styles are a step toward greater control, but whether Claude can fully shed its cheerleading roots remains to be seen.
