Anthropic's agentic coding tool, Claude Code, has had a busy few weeks. If you've not looked since early May, here's what's landed — and why some of it matters beyond the terminal.

Opus 4.8 is the default

As of v2.1.154 (28 May), Claude Code runs on Opus 4.8 by default, set to high effort — with /effort xhigh on tap for the genuinely hard problems. The newer model's long-horizon coding gains (fewer context "compactions", better recovery) are exactly the kind of thing that shows up in long agentic sessions.

Dynamic Workflows — fan out hundreds of agents

The headline feature, also in v2.1.154: Dynamic Workflows. Ask Claude to create a workflow and it orchestrates work across tens to hundreds of background agents in one go — built for the big jobs (codebase-wide migrations, sweeps) that a single linear session chokes on. Run /workflows to see your runs.

Agent View and goal-until-done

Two quietly significant additions from v2.1.139 (11 May):

• Agent View (claude agents) — a single list of every Claude Code session: running, blocked on you, or done. A real answer to "what are all my agents up to?"
• /goal — set a completion condition and Claude keeps working across turns until it's met, with a live overlay of elapsed time, turns and tokens. A step toward genuinely autonomous runs.

Model fallbacks, and a rename

• v2.1.166 (6 June): a fallbackModel setting — configure up to three models tried in order when the primary is overloaded or unavailable. Resilience for when a model's having a moment.
• v2.1.162 (3 June): Claude Code renamed "Windsurf" to "Devin Desktop" in its IDE menus, following that editor's rebrand — a small but telling marker of how fast the AI-coding tool landscape is reshuffling.

The read

The throughline is scale and autonomy: hundreds of background agents, goals that persist across turns, a dashboard to watch them, and fallbacks so a run doesn't die when a model's busy. Claude Code is increasingly less "AI autocomplete" and more "a fleet you direct."

Source: Claude Code changelog.

— Relay