Claude Code Ultracode: Full Power Mode for Parallel Agents

I spend most of my day inside Claude Code. It is the closest thing I have found to a pair programmer who actually reads the codebase before typing. So when Anthropic dropped a new /effort option called ultracode alongside Opus 4.8, I immediately opened a fresh terminal to see what changed.

The short version: ultracode is Claude Code's "Full Power Mode." It combines the deepest reasoning level (xhigh) with a new internal feature called Dynamic Workflows that turns a single Claude Code session into a self-managing team of parallel agents. Instead of one model handling one task in one conversation, Claude breaks the work down, spawns sub-agents, runs them in parallel, and has another agent verify and refute the results until the answer converges.

This is not a small tweak. It is a shift in how agentic coding works at the terminal.

What Ultracode Actually Does

When you select ultracode from the /effort menu, two things happen under the hood:

1. xhigh reasoning - the model runs at the deepest reasoning tier, which means longer thinking time but significantly better planning and debugging.
2. Dynamic Workflows - Claude Code itself decides whether a task is complex enough to warrant parallelization. If it is, the system automatically forms a team: one agent plans, multiple agents execute in parallel, and a verifier agent checks the output. The loop repeats until the result is solid.

The key insight is that this is a Claude Code-side setting, not a model API change. You are not calling a new ultracode parameter in the Anthropic API. Instead, Claude Code internally routes the request as xhigh and then orchestrates the workflow agents itself. That means the feature is tightly coupled to the CLI tool, not the underlying model endpoint.

Also, the setting is session-specific. Start a new session and you are back to the default effort level. You must re-run /effort and pick ultracode every time you want it.

When to Use It

Ultracode is designed for heavy tasks that benefit from parallel investigation and verification. The kinds of work where I would actually use it:

• Security audits across the entire codebase - checking authentication flows, input validation, and dangerous patterns in every file, not just the one I am looking at.
• Large-scale refactoring - migrations touching hundreds of thousands of lines where I need existing tests to keep passing while the structure changes.
• Research tasks - investigating a topic from multiple angles, verifying facts, and returning a report with citations.

Conversely, it is overkill for quick fixes and simple questions. If I am changing a single config value or asking about a function signature, ultracode adds latency without value. The overhead of team formation and verification only pays off when the task is genuinely large.

Prerequisites

Before ultracode shows up in your /effort menu, three things must be true:

I ran into the plan issue myself. On my work machine, where I have a Team plan, ultracode appeared immediately. On my personal Pro account, it did not show up until I went into /config, found the "Dynamic workflows" line, and flipped it to true. If you meet the model and version requirements but still do not see ultracode, check /config before assuming your plan is unsupported.

How to Enable It

Once the prerequisites are met, activation is three steps:

# Step 1: Open /config and toggle "Dynamic workflows" to true
# Step 2: Switch to Opus 4.8
/model Opus 4.8

# Step 3: Open /effort and select ultracode
/effort
# choose ultracode from the list

That is it. The setting is active for the current session and you will see Claude Code parallelize work when it judges the task to be complex enough.

What This Means for Agentic Workflows

Ultracode is one of the clearest signals that agentic coding is moving from "one smart assistant" to "a team of specialists." The model is no longer just answering questions. It is deciding whether to delegate, how to split the work, and how to verify the results.

For developers who already live in Claude Code, this is a genuine productivity upgrade for the right tasks. For everyone else, it is worth watching because it sets the pattern for how CLI-based AI tools will evolve: tighter integration between reasoning depth and workflow orchestration, all inside the terminal.

I will be running ultracode on my next large refactor and comparing the output quality against the same task with high effort. If the parallel verification actually catches edge cases I would have missed, it will become my default for any task that touches more than a few files.

Source: Tokyo PC Rescue Captain - note.com