✦ Claude Code /loop — Turn Your Session into an Autonomous Background Worker

Introduced in Claude Code v2.1.71 (early March 2026), the /loop command is a session-scoped cron scheduler that lets you give Claude a prompt and a cadence and have it repeat automatically — no manual re-triggering required. Closing the terminal cancels everything; this is deliberately a lightweight, session-bound tool, not a persistent background daemon. The underlying primitives are three native tools exposed to skills: CronCreate, CronList, and CronDelete.

Basic syntax

/loop 5m check the deployment and report any errors
/loop 30m run the test suite and summarise failures
/loop 1h scan open pull requests and flag stale ones

Supported interval units are s (seconds), m (minutes), h (hours), and d (days). Intervals below one minute are rounded up since the underlying cron has one-minute granularity. Each scheduled task receives an 8-character ID that you can use with CronDelete to cancel it individually.

Key constraints to know

• Session-only — tasks are cancelled when Claude Code exits. Restarting the session clears the task queue.
• Idle-gated — a task only fires when Claude Code is idle; if you are mid-conversation, the next execution waits until you are done.
• 3-day expiry — recurring tasks automatically expire 72 hours after creation, fire one final time, then delete themselves. This prevents forgotten loops from running indefinitely.
• 50-task limit — a session can hold at most 50 scheduled tasks simultaneously.

Practical use cases

• Monitor a CI/CD deploy every 5 minutes and notify you in-session when it goes green.
• Scan an error log every hour overnight and file a pull request if a new pattern is detected.
• Summarise overnight commit activity every morning before you start work.
• Watch for deprecated API patterns in a growing codebase on a daily cadence.

✦ Claude Marketplace — Anthropic's Enterprise App Store (No Commission)

On 6 March 2026, Anthropic launched the Claude Marketplace in limited preview — a B2B procurement platform that lets enterprises with existing Anthropic spending commitments buy Claude-powered software from vetted third-party partners, all within a single billing relationship. Crucially, Anthropic takes no commission on marketplace transactions, positioning it as a procurement convenience rather than a revenue share play.

Launch partners

• GitLab — AI-assisted DevSecOps workflows
• Harvey — legal contract drafting and analysis
• Lovable — AI-accelerated front-end development
• Replit — cloud-native coding and deployment
• Rogo — financial data analysis for investment professionals
• Snowflake — AI-powered data cloud integrations

How procurement works

Enterprises that have signed annual API spend commitments with Anthropic can apply a portion of that commitment to marketplace purchases — so the cost flows through a single vendor relationship rather than requiring separate contracts with each ISV. Payment is handled by Anthropic; the partner receives their revenue directly. Organisations without an existing spend commitment can still browse, but must negotiate separately.

✦ Claude's Business Subscription Share Jumps 4.9 % in a Month

Data published by fintech company Ramp on 19 March 2026 shows that Anthropic's business software subscriptions grew 4.9 % month-over-month in February 2026 — the same period in which OpenAI's share fell 1.5 %. The headline numbers: OpenAI still leads at 34.4 % of business AI subscriptions tracked by Ramp, but Anthropic has climbed to 24.4 %, up from roughly 4 % just twelve months earlier.

The speed of the shift

• In January 2026 alone, Anthropic gained 2.8 percentage points while OpenAI lost 0.9.
• "Nearly one in four businesses on Ramp now pays for Anthropic; a year ago it was one in 25."
• Among companies buying AI for the first time, Anthropic wins approximately 70 % of head-to-head decisions against OpenAI.

Revenue trajectory

Anthropic's annualised revenue grew from $1 billion in December 2024 to $4 billion by July 2025, $9 billion by December 2025, and $14 billion by February 2026 — a roughly 14× increase in 14 months. The company now serves more than 300,000 business customers.

✦ Fixing Claude with Claude — Anthropic's AI Reliability Engineering in Practice

On 19 March 2026, a member of Anthropic's AI Reliability Engineering (AIRE) team spoke at QCon London and published findings on using Claude as an internal SRE tool — in effect, using the model to help maintain the infrastructure that runs the model. The report is a candid account of both the genuine productivity gains and the current limitations that prevent Claude from fully replacing a human site reliability engineer.

What AIRE does

Anthropic's AI Reliability Engineering team partners with internal teams to improve reliability across every serving path: from the SDK through network and API layers, to serving infrastructure and accelerators. Claude agents assist by monitoring dashboards, correlating incidents, drafting runbooks, and proposing mitigations — all faster than a human on-call engineer could scan the same signal volume.

Where Claude excels

• Pattern recognition at scale — scanning large volumes of logs and metrics to surface anomalies a human might miss in the noise.
• Runbook drafting — given an incident description, Claude rapidly drafts a structured runbook that on-call engineers then validate and execute.
• Incident summarisation — compressing multi-hour incident timelines into clear post-mortem summaries.

Where Claude still falls short

• Correlation vs. causation — the AIRE team explicitly called out Claude's tendency to mistake correlated signals for root causes, which can send engineers down the wrong diagnostic path.
• Novel failure modes — Claude is effective on patterns seen in training but less reliable on genuinely new infrastructure failure classes.
• Autonomous remediation — all mitigations still require human approval; fully autonomous incident resolution remains out of scope for now.