Descriptions:
Craig Hewitt shares a detailed firsthand assessment of Anthropic’s Claude Opus 4.7 after a day and a half of intensive testing across knowledge work and agentic coding workflows. The video is particularly relevant for Claude Code and Claude Co-work users who noticed disrupted results immediately after the model upgrade.
The central finding is that Opus 4.7 is far more literal and deterministic than its predecessor, Opus 4.6. Where 4.6 could fill gaps in vague or unstructured prompts — a strength Hewitt relied on heavily for marketing analysis, financial reports, and content creation — 4.7 demands explicit, well-constrained inputs or it underdelivers. Hewitt identifies four additional behavioral shifts: the model defaults to “extra high effort” reasoning, consuming tokens aggressively; it no longer spawns parallel sub-agents automatically, requiring explicit prompting to avoid context window bloat; the thinking-budget cap mechanism from 4.6 is gone, replaced by fully adaptive reasoning; and image analysis and presentation generation are meaningfully improved.
Hewitt’s practical recommendation is a deliberate three-tier model routing strategy: use Opus 4.7 for structured, long-running agentic coding tasks with detailed prompts; use Sonnet for knowledge work and lighter cognitive tasks; and retain Opus 4.6 for workflows where the newer model still underperforms. He also highlights the Superpowers plugin for sub-agent orchestration as worth watching. The video is essential context for any team that has built business workflows on Claude and needs to understand what changed in 4.7 — and how to adapt accordingly.
📺 Source: Craig Hewitt · Published April 18, 2026
🏷️ Format: Review
