$ recombobulate _
home / tips / use-max-effort-for-architectural-decisions
0

Use Max Effort for Architectural Decisions

bagwaa @bagwaa · Mar 26, 2026 · Performance
use-max-effort-for-architectural-decisions

Not every task deserves the same level of thinking. For a quick rename or a simple bug fix, Claude Code's default reasoning is fine. But when you're making structural decisions that will ripple through your entire codebase, you want Claude working at full capacity.

Claude Code supports different effort levels. For architectural work, pushing to the maximum level produces noticeably better output compared to the standard high setting.

# When asking Claude to design something significant, be explicit:
"Use maximum effort to design the authentication architecture for this app."

# Or lean on the ultrathink keyword:
"ultrathink: what's the best way to structure our data access layer?"

The tradeoff is real: max effort burns through tokens significantly faster than high effort. Use it strategically, not for every task.

Good candidates for max effort:

  • Designing system architecture or data models
  • Reviewing complex, interconnected code before a refactor
  • Evaluating tradeoffs between different implementation approaches
  • Debugging a gnarly issue with unclear root cause

For routine coding tasks, file edits, or anything straightforward, the standard effort level is faster and cheaper.

Think of max effort like calling in a senior architect rather than a capable developer. Right tool, right moment.


via @rayyzinoo

~/recombobulate $ tip --comments --count=0

Log in to leave a comment.

~/recombobulate $ tip --related --limit=3
0
Filter Test Output with a PreToolUse Hook to Cut Token Costs

A PreToolUse hook can intercept test runner commands and filter output to show only failures, cutting thousands of tokens from Claude's context.

bagwaa @bagwaa · 3 hours ago
0
Move Specialised CLAUDE.md Instructions into Skills to Shrink Context

CLAUDE.md loads into every message. Move workflow-specific instructions into skills that load on demand to reduce token costs across your session.

bagwaa @bagwaa · 3 hours ago
0
Use prompt.id to Trace All Activity from a Single User Prompt

Every event emitted while processing a single prompt shares a prompt.id UUID, letting you trace the complete chain of API calls and tool executions.

bagwaa @bagwaa · 3 hours ago