Evolving Claude APIs for Agents

Time: 8:49 AM (approx)

Speaker Bio: Head of API Engineering at Anthropic. Previously worked at Stripe and Betterment. Leading the Claude Developer Platform.

Speaker Profile: Full Speaker Profile

Company: Anthropic develops Claude, with capabilities including parallel tool execution, Model Context Protocol (MCP), and extended thinking with tool use.

Focus: Understand how Anthropic’s latest Claude APIs enable developers to build “unhobbled” autonomous agents that can work for hours on complex tasks. Pay attention to MCP and context management features.

Slides

Slide: 08-49

Key Point: This is the opening keynote slide introducing the conference theme about AI’s role in the future of software development.

Literal Content:

• Dark blue background with white text
• “CODE SUMMIT” header
• “Keynote: The Future of AI in Code”
• Network diagram icon with code brackets in center
• Speaker: “Duane Moore”
• Organization: “Jemaith and in Islett”

Slide: 08-50

Key Point: Showcases the major corporate sponsors and technology partners supporting the AI Engineering Code Summit, representing diverse industries from finance to tech to consulting.

Literal Content:

• Brown/tan background
• Company logos including: Cerebras, ux+, workato, OpenAI, McKinsey & Company
• DATADOG, AWS, gimlet, DX, Capital One
• Northwestern Mutual, OpenHands, HumanLayer, METR
• qodo, cline, manus, MINIMAX, NETFLIX
• Temporal, Applied Compute (highlighted with red box), Bloomberg, Robinhood
• Tenex, PREFECT

Slide: 09-08

Key Point: Introduces the three core pillars of Anthropic’s Claude Developer Platform: leveraging Claude’s AI capabilities, managing context effectively, and providing Claude with computer access for autonomous work.

Literal Content:

• Pink/rose background
• Title: “How the Claude Developer Platform maximizes intelligence”
• Three boxes with icons:
  1. Lightbulb icon - “Harness Claude’s capabilities”
  2. Magnifying glass/network icon - “Manage Claude’s context”
  3. Code window icon - “Give Claude a computer & let it cook”
• “ANTHROPIC” footer with page number 3

Notes

• Building powerful agentic systems using claude
• “Give claude a computer”
• Claude running tools
  • How does this fit in with skills
• Managing context window
• MCP as a way to interact with external systems
• Memory tool as a way to bring back info into the context when you need it
  • First version is the file system
  • https://code.claude.com/docs/en/memory
• Context editing
  • Helps you clear things out that shouldn’t be in the window
  • Clearing out tool results
• Memory tool with context editing -> the way to go
• Code execution tool
  • Can we do this with dagger

Anthropic is hiring

Key Points Summary

Core Thesis: Building powerful agentic systems by giving Claude computer access and the right tools

Main Components:

1. Tool Integration - Claude running tools directly, with Skills as a framework for specialized capabilities
2. MCP (Model Context Protocol) - Standard way to connect Claude to external systems and data sources
3. Memory System - Retrieves relevant context when needed (file system-based initially)
4. Context Editing - Manually clear unnecessary content (especially old tool results) from the window
5. Best Practice: Memory tool + context editing together = optimal context management
6. Code Execution - Native tool (question: Can Dagger be used here?)

Bottom Line: Effective agentic systems require both expanding capabilities (tools, MCP) and managing constraints (memory, context editing)

Tweet Options

Concise version: Building powerful agentic systems with Claude: Give it computer access, connect external systems via MCP, manage context with memory + editing, and let it execute code. The key is balancing capability expansion with smart context management.

Technical version: Claude agentic system stack: • Tools + Skills for capabilities • MCP for external integrations • Memory tool for context retrieval • Context editing to clear bloat Combo of memory + editing = the way to manage your context window effectively

Punchy version: Want powerful Claude agents? Give Claude a computer, connect it to your systems (MCP), manage the context window smartly (memory + editing), and let it execute code. Context management is just as critical as capability expansion.

Which style fits your audience best?