🧭 Anthropic's IPO Roadshow Opens — Goldman and Morgan Stanley Kick Off Two-Week Investor Tour

Three days after filing its S-1, Anthropic has officially opened its institutional investor roadshow for the planned November Nasdaq listing. Goldman Sachs and Morgan Stanley are leading the underwriting syndicate, with Citigroup, JPMorgan, and Allen & Company rounding out the bookrunning group. The roadshow is scheduled across 14 cities in 12 days, with New York, San Francisco, London, Singapore, and Tokyo as the anchor stops. A public webcast version is available to retail investors through the SEC's EDGAR system — a requirement since the JOBS Act Modernization Act of 2024 mandated equal-access roadshows for Tier 1 AI companies.

The pitch narrative

• Revenue trajectory: $14.2B H1 2026 revenue, $28–30B annualised run-rate heading into the listing. Enterprise accounts (≥$1M ARR) now represent 61% of revenue, up from 44% a year ago.
• Path to profitability: First adjusted-profitable quarter (Q3 2026) cited as evidence that the model training cost curve has bent. Gross margins improved from 38% to 54% year-over-year as inference efficiency gains compounded.
• Moat claims: Constitutional AI, the Model Welfare Commitments, and the METR safety evaluation arrangement are positioned as structural differentiators — not CSR, but as inputs to enterprise risk management that competitors cannot quickly replicate.
• Capital use: Proceeds earmarked for next-generation model training infrastructure (dedicated data centres in Oregon and Virginia), geographic expansion (EU AI Act compliance infrastructure), and the Agentic AI Foundation governance work.

🧭 Claude Code v2.1.310: Hooks Debugger and /doctor Environment Check

Claude Code v2.1.310 ships two features that directly address the two most common support requests in the community: "why is my hook not firing?" and "why does Claude Code behave differently on my CI box vs. my laptop?" The new --hooks-debug flag and the new /doctor slash command tackle these head-on.

--hooks-debug flag

Append --hooks-debug to any Claude Code invocation to see a real-time trace of every hook lifecycle event:

claude --hooks-debug "fix the auth bug"

[hooks] PreToolUse → bash_command matched by rule "require-approval"
[hooks] PreToolUse → exit_code=0, output="approved", continue=true
[hooks] PostToolUse → bash_command fired 1 hook(s)
[hooks] Stop → no matching rules

The trace includes the matched rule name, exit code, stdout output, and the continue / block decision made. This replaces the previous approach of adding echo statements inside hook scripts to understand what was happening.

/doctor command

Running /doctor inside a session produces a structured environment report:

• Claude Code version and whether an update is available
• settings.json load path (which file is actually being read — global, project, or none)
• Active hooks — lists all loaded hook rules with their event types and match patterns
• MCP server status — each configured server's connection state and last ping time
• Sandbox mode — whether the Linux/macOS sandbox is active and which deny rules are loaded
• Model in use — current default model and whether Fast Mode is active
• API key source — env var, keychain, or .claude config, without revealing the key value

Other changes in v2.1.310

• Hook error surfacing: When a hook exits with a non-zero code and continue=false, the blocking reason is now shown inline in the tool-use stream (not just in the session log), so developers see immediately which rule blocked an action.
• Worktree isolation fix: EnterWorktree now correctly inherits the parent session's MCP configuration, fixing a regression introduced in v2.1.307.
• Performance: Settings file parsing is now cached per session — previously re-parsed on every tool call, adding ~4 ms overhead in repos with large settings.json files.

🧭 Claude.ai Projects 2.0 — Real-Time Collaborative Workspaces for Teams

Anthropic has released Projects 2.0 for Claude.ai Team and Enterprise plans, transforming Projects from a personal context container into a fully collaborative, real-time shared workspace. The key changes: multiple team members can now contribute to a Project's knowledge base simultaneously, conversations are shared and replayable, and a new Project Memory layer gives the team a structured way to define persistent facts and preferences that Claude applies across every conversation in the Project — without each person having to re-explain context in their individual sessions.

What's new in Projects 2.0

• Shared knowledge base: Any team member with edit access can add documents, code, data files, or custom instructions to the Project. Changes propagate to all active sessions within seconds. Previously, each user maintained a personal copy.
• Conversation sharing and replay: Conversations can be marked as shared, making them visible (read-only) to the whole team. Managers can review what questions were asked and what Claude suggested — useful for compliance workflows and knowledge capture.
• Project Memory: A structured key-value store (up to 50 entries) that persists across all conversations. Teams define things like "our API base URL is…", "we follow this naming convention…", or "always cite sources with this format". Claude reads Project Memory at the start of every session.
• Role-based access: Owner → Admin → Editor → Viewer roles, integrated with the organisation's SSO groups. IT administrators can now provision Project access via SCIM.
• Usage attribution: The admin dashboard now breaks down token consumption by Project, by user within a Project, and by conversation — making it easier to chargeback AI usage to specific teams or client engagements.