What Is Code with Claude 2026? Anthropic's Dev Conference
What Is Code with Claude 2026? Anthropic’s Dev Conference
Code with Claude is Anthropic’s developer event series. The flagship San Francisco event in May 2026 delivered three major announcements: Managed Agents (Anthropic-hosted agent runtime), Proactive Workflows (event-driven agents), and the Capability Curve (Anthropic’s framing for how agent autonomy will compound through 2028). Here’s what was announced and what it means for developers.
Last verified: May 23, 2026
The 30-second summary
| Details | |
|---|---|
| Event | Code with Claude |
| Vendor | Anthropic |
| May 2026 location | San Francisco (flagship), with regional events in London (May 20-21) and Tokyo (June 5-6) |
| Headline announcements | Managed Agents, Proactive Workflows, Capability Curve |
| Featured customers | Cursor, GitHub, Asana, Replit, Vercel |
| Keynote | Dario Amodei (CEO, Anthropic) |
| Audience | Developers, technical founders, enterprise AI teams |
| Source | InfoQ coverage of the SF event |
What was announced
1. Managed Agents — Anthropic-hosted agent runtime
The biggest practical announcement. Managed Agents is an Anthropic-hosted runtime for long-running Claude agents — you bring the agent definition (system prompt, tools, MCP connections), Anthropic runs the infrastructure.
What you get out of the box:
- Long-running execution. Agents can run for hours or days with state preservation.
- Hosted MCP connections. Anthropic manages credentials, retries, and reliability for MCP servers.
- Built-in observability. Logs, traces, cost tracking, and a UI to inspect agent runs.
- Outcome-based pricing (in preview). Pay per successful outcome, not per token.
- Enterprise features. SSO, RBAC, audit logs, SOC 2.
The trade-off vs the Claude Agent SDK: less control, faster to production. Managed Agents is for teams that don’t want to operate agent infrastructure themselves. The SDK is for teams that want full control of the runtime.
(Deeper dive on Managed Agents.)
2. Proactive Workflows — event-driven agents
A new pattern: agents that trigger themselves based on events, not just user prompts.
Examples Anthropic demoed:
- A “watch our GitHub repo and respond to security advisories” agent.
- A “monitor our Stripe dashboard and alert when refunds spike” agent.
- A “process incoming sales emails and route to the right team” agent.
The technical primitive: MCP event sources that fire and wake up a Claude agent with context. The agent runs, takes action (or asks for approval), and goes back to sleep.
This is functionally similar to AWS Lambda triggers, but for AI agents instead of code. The shape of agent products will shift through Q3-Q4 2026 as developers build event-driven rather than chat-driven agent UX.
3. The Capability Curve — strategic framing
Anthropic’s framing for how much economic work an agent can do autonomously, measured by the duration of a task that requires no human intervention.
| Year | Agent autonomy ceiling (typical) |
|---|---|
| 2024 | Minutes (single chat turn, one tool call) |
| 2025 | Tens of minutes (chained tool calls, simple workflows) |
| May 2026 | 1-4 hours (multi-file refactors, multi-step research) |
| 2027 (projected) | Multi-day tasks (run a deal pipeline, manage a project) |
| 2028 (projected) | Week-long tasks (run a small business function autonomously) |
The Capability Curve framing matters because it tells developers what’s worth investing in now:
- Tasks under 1-4 hours: automate today with Claude Opus 4.7 + Agent SDK.
- Tasks at the boundary (4-8 hours): build prototypes, expect to need human checkpoints, plan for 2027 fully-autonomous deployment.
- Multi-day tasks: research and design only; don’t try to ship in 2026.
Dario also reiterated his earlier prediction that a one-person billion-dollar company will emerge in 2026, noting that two-person AI-augmented companies have already crossed a billion dollars in valuation.
Featured customer talks
Anthropic always uses Code with Claude to showcase how customers ship Claude-powered products. May 2026 lineup:
- Cursor — How Composer 2.5 and Cursor’s agent system orchestrate Claude Opus 4.7 for hard refactors.
- GitHub — Copilot’s integration with Claude for code review and PR generation.
- Asana — Claude-powered workflow automation across Asana’s project graph.
- Replit — Replit Agent shipping production apps with Claude as the planner.
- Vercel — V0 and Vercel’s agent surfaces using Claude for full-stack generation.
The customer talks are the most useful part of Code with Claude for working developers — concrete production patterns from teams shipping at scale.
What Code with Claude is not
- Not Anthropic’s general developer event — there are also Anthropic-hosted enterprise webinars (legal teams, government, healthcare) and the broader Code with Claude conference circuit.
- Not free — the flagship SF event has paid attendance, though Anthropic streams keynotes publicly.
- Not as broad as AWS re:Invent or Google I/O — it’s developer-focused, single-vendor, and product-launch-driven.
How Code with Claude compares to similar events
| Event | Vendor | Audience | Scale |
|---|---|---|---|
| Code with Claude (SF) | Anthropic | Developers | ~2,000-3,000 |
| OpenAI DevDay | OpenAI | Developers | ~2,500 (2024) |
| Google I/O | Developers + consumers | 5,000+ in-person, millions online | |
| Microsoft Build | Microsoft | Developers + enterprise | ~6,000 in-person |
| AWS re:Invent | AWS | Cloud + AI developers | 50,000+ |
| NVIDIA GTC | NVIDIA | AI researchers + developers | 20,000+ in-person |
Code with Claude is smaller and more focused than the big platform events. The trade-off: less spectacle, more substance. If you build on Claude, the May 2026 announcements (Managed Agents, Proactive Workflows) are directly actionable.
What to take away as a developer
1. Managed Agents is worth piloting. If you’re running production Claude agents on your own infrastructure today, Managed Agents probably saves you 80% of the ops work. Pilot one workflow.
2. Build event-driven agents. Proactive Workflows is a real pattern shift. Your next agent shouldn’t be a chat UI — it should trigger on events and act on its own.
3. Time your roadmap to the Capability Curve. Don’t try to ship multi-day autonomous agents in 2026 — they’ll break. Do ship multi-hour agents today. Plan for multi-day autonomy in 2027.
4. Watch for follow-up announcements at Code with Claude London (May 20-21) and Tokyo (June 5-6). Regional events often include EU/APAC-specific feature announcements (data residency, language model variants, regional MCP connectors).
Verdict
Code with Claude 2026 was Anthropic’s most ambitious developer event to date — Managed Agents is a meaningful product launch, Proactive Workflows is a pattern that will shape agent UX through 2027, and the Capability Curve framing gives developers a way to time investments.
If you build on Claude: pilot Managed Agents, design for event-driven workflows, and don’t overshoot the Capability Curve.