Managed Agents — Enterprise GA and “Dreaming” #
- Anthropic scales up with enterprise features for Claude Cowork and Managed Agents (9to5Mac, 2026-04-09) — Managed Agents reach enterprise GA: sandbox support, private MCP servers, role-based access control, group spend limits, expanded OpenTelemetry. The move from beta to GA is the enterprise-readiness signal for managed agentic workflows.
- Anthropic’s Code with Claude: Managed Agents, Proactive Workflows, Capability Curve (InfoQ, 2026-05) — May 2026 Code with Claude event: multi-agent orchestration, Outcomes feature, and “Dreaming” — Claude inspects its own past sessions to identify patterns and self-improve without model retraining. The most architecturally significant new capability: the boundary between model capability and tool capability is blurring.
- Anthropic rolls out AI agents for financial services (TechRadar) — 10 prebuilt financial services agents deployable in days via Cowork, Claude Code, and Managed Agents. Vertical-specific agents compete on deployment simplicity, not raw capability.
Agentic Coding at Scale — Primary Data #
- 2026 Agentic Coding Trends Report (Anthropic) — API volume up 17× year-on-year; data from Anthropic’s own usage telemetry on agentic coding patterns. Primary source from the platform itself.
Workflow Patterns — Community and Practitioner #
- Claude Code Tips I Wish I’d Had From Day One (Marmelab) — Plan mode first;
/rewind and /clear for recovery; commit working state before escalating prompts. Practical workflow safety patterns from an engineering consultancy. - claude-code-tips (ykdojo) (GitHub) — 45 tips including: using Gemini CLI as Claude Code’s assistant (“minion pattern”), halving system prompt size, running Claude Code inside a container. The Gemini-as-minion pattern is the most novel — a second-tier model for lightweight tasks while Claude handles complex reasoning.
- A New Way to Extract Detailed Transcripts from Claude Code (Simon Willison, Substack) — New technique for extracting detailed session transcripts. Practical meta-tooling for audit, review, and debugging of agentic sessions.
- An Update on Recent Claude Code Quality Reports (Simon Willison, 2026-04-24) — Willison’s analysis of community reports about Claude Code quality regressions. The quality-vs-capability narrative is separate from the security story (last gather) — worth tracking as an ongoing thread.
Cross-links #
- Vibe Coding Approaches The “Dreaming” feature (Claude Code learning from past sessions) is the closest thing to persistent skill accumulation in a mainstream coding tool — distinct from model updates. Directly relevant to the agentic governance question in the vibe-coding entry.
- Claude Integrations Managed Agents enterprise GA (private MCP servers, role-based access, OpenTelemetry) is the infrastructure that makes the KPMG 276K and PwC global deployments possible.
- Applications of Vibe Coding Financial services vertical agents (TechRadar) are a concrete instance of the “citizen developer within a governed sandbox” model — prebuilt agents with guardrails, not open-ended generation.
- Quality signal: “Dreaming” — Claude Code inspecting its own session history to self-improve without model retraining — is the first instance of agentic self-improvement in a mainstream coding tool. Architecturally significant: model capability and tool capability are no longer cleanly separated.
- Emerging theme: The Gemini-as-minion pattern (ykdojo) suggests the multi-model workflow is hardening into practitioner norm: cheaper/faster models for lightweight tasks, Claude for complex reasoning. Changes cost-optimisation thinking in agentic setups.
- Keyword suggestion:
"claude dreaming" self-improvement session history — new enough that coverage is sparse; tracking rollout and community response is worthwhile. - Method note: The Anthropic Agentic Coding Trends Report PDF (resources.anthropic.com/hubfs/) is a primary data source. Check for updated versions at each gather cycle.