What's new in Claude Code in April 2026?

Self-serve Enterprise plans (no Sales call required), unified seat type with Claude + Claude Code + Cowork access, an OAuth token reload fix, MCP plugin reload improvements, and a Bun NO_PROXY fix. Plus better governance hooks for team admins.

Do I still need to talk to sales for an Enterprise plan?

No. As of April 2026 you can purchase an Enterprise plan directly from Anthropic's website. Self-serve Enterprise includes a single seat type that bundles Claude, Claude Code, and Cowork access — no Sales conversation required.

What is Cowork in the Claude Code bundle?

Cowork is Anthropic's collaborative workspace surface for shared agent sessions, project context, and team-level prompt and skill libraries. It's now bundled into the Enterprise seat alongside Claude and Claude Code.

How is Claude Code being used at scale?

Stripe deployed Claude Code to 1,370 engineers via a zero-configuration enterprise binary. That deployment is the public benchmark for what 'Claude Code at scale' looks like in 2026.

Quick Answer

Claude Code Enterprise Governance: What Changed (April 2026)

Published: April 26, 2026

Claude Code Enterprise Governance: What Changed (April 2026)

Late April 2026 saw Anthropic ship its biggest Claude Code enterprise update in months — quietly, with no big announcement. If you run Claude Code across an engineering team, here’s what changed and why it matters.

Last verified: April 26, 2026

The big one: self-serve Enterprise plans

Previously, Anthropic Enterprise plans required a Sales conversation. As of April 2026:

Any organization can buy Enterprise directly on anthropic.com — no Sales call
One seat type bundles Claude (chat), Claude Code (CLI/IDE), and Cowork (collaborative workspace)
Provisioning, SSO, and audit logs are configurable from the admin console
SCIM and SSO providers (Okta, Azure AD, Google Workspace) supported

This matters because it removes the single biggest friction point for mid-market adoption — small companies with 50–500 engineers no longer have to go through a 4-week procurement cycle to get a unified license.

What’s in the Cowork bundle

Cowork is Anthropic’s team workspace — bundled into Enterprise seats now. It includes:

Shared agent sessions — multiple humans observe and intervene in one Claude Code run
Team skills/prompt library — versioned, reviewed, deployed across the org
Project context store — long-lived context bundles per repo/project
Audit trail — who ran what, when, with what tools

If your team has been hand-rolling shared prompts in Notion, Cowork is the official Anthropic answer.

Recent Claude Code release fixes (last 30 days)

From the public Anthropic release notes:

Fix	Why it matters
`/login` no longer breaks when launched with `CLAUDE_CODE_OAUTH_TOKEN` after token expiry	Long-running CI/agent setups stop dying silently
Prompt-input undo (`Ctrl+_`) now works immediately after typing	Quality-of-life in the TUI
`NO_PROXY` respected for remote API requests under Bun	Corporate network compatibility
Spurious escape/return triggers fixed for slow connections	SSH / remote pairing now reliable
SDK `reload_plugins` no longer reconnects all MCP servers serially	Faster MCP plugin reload — feels instant now

These are paper-cut fixes individually; together they reflect Claude Code maturing into a tool people actually leave running for hours.

Stripe deployment: the public benchmark

Anthropic publicly highlights Stripe’s Claude Code deployment to 1,370 engineers via a “zero-configuration enterprise binary.” That’s the largest publicly-confirmed Claude Code deployment, and it sets the bar for what “enterprise scale” looks like.

Key elements of the Stripe pattern (as publicly described):

Zero-config binary — engineers install one CLI, no per-machine setup
Centrally managed model selection and tool allowlists
No Anthropic servers touched for sensitive code paths (where governance matters)
Per-team observability — usage, model mix, tool calls

Most Claude Code Enterprise deployments will resemble some subset of this pattern in 2026.

Governance features now standard

If you adopt Claude Code Enterprise in April 2026, you get out-of-the-box:

SSO + SCIM provisioning (Okta, Azure AD, Google Workspace, OneLogin)
Per-team model allowlists (e.g., “Sonnet only” for one team, “Opus + Sonnet” for another)
MCP server allowlists — control which tools each team can connect to
Skill library governance — review and approve shared agent skills
Token-budget controls per team and per user
Audit logs with API export
Data residency options via Bedrock or Vertex AI deployments

This closes the gap with GitHub Copilot Enterprise on governance — historically the main reason large orgs preferred Copilot over Claude Code despite the model quality difference.

Pricing reality check

Self-serve Enterprise pricing as of April 2026 is publicly listed for the first time. Confirm current pricing on the Anthropic site, but the structure is:

Per-seat monthly for the bundled Claude + Claude Code + Cowork
Tiered usage caps with overage pricing
Annual contract discount still requires Sales engagement for the largest deals

For teams of 20–200 engineers, the self-serve path is now the fastest way to consolidate ChatGPT Team + Cursor + a coding-agent SaaS into one Anthropic bundle.

What this means for your stack

If you already use Claude Code informally

Upgrade to Enterprise self-serve. You get SSO, audit, and centralized model selection without the Sales-call overhead. The 30-day Enterprise trial path is now standard.

If you use GitHub Copilot Enterprise

Pilot Claude Code on one team. Copilot still wins on Visual Studio depth and IDE autocomplete polish, but Claude Code Opus 4.7 leads SWE-bench Verified and now has the governance story to match.

If you use Cursor

Claude Code is complementary, not a replacement. Cursor remains the best chat-driven IDE. Claude Code is the best autonomous CLI agent. Many teams run both: Cursor in the editor, Claude Code in the terminal for long tasks.

If you build agents with Claude (not for coding)

The April 2026 SDK improvements — especially the reload_plugins fix — make Claude as an agent runtime more reliable. MCP integrations no longer take a perceptible hit when reloading.

What’s still missing

To be balanced, here’s what Claude Code Enterprise still doesn’t have as of April 2026:

Native VS Code extension parity — Cursor and the Anthropic VS Code extension still have rougher edges than the JetBrains/CLI experience
First-party fine-tuning UI for Code — coming via Vertex AI / Bedrock but not native
Granular per-file governance — you can allowlist tools, but per-file write controls require external policy layers

Bottom line

April 2026 is the month Claude Code grew up as an enterprise product. Self-serve Enterprise plans, the Cowork bundle, governance-grade SSO/SCIM, and publicly-validated deployments at Stripe-scale together make this the cleanest moment so far to pick Claude Code as your standard coding agent — without the procurement overhead that defined 2025.

Last verified: April 26, 2026. Sources: Anthropic release notes (releasebot.io/updates/anthropic), code.claude.com/docs/en/overview, Anthropic.com/product/claude-code (Stripe case study), public Reddit reporting on April 2026 enterprise rollout.