Pages tagged "infrastructure"

9 pages tagged with infrastructure. ← all tags

• Agent Payment Protocols — The Infrastructure Under Agentic Commerce Four standards shipped in 2025-2026 define how AI agents transact: AP2 (Google's universal payments rail, September 2025), x402 (Coinbase/Cloudflare HTTP-native stablecoin protocol), UCP (Universal Commerce Protocol — the commerce schema), and Visa TAP (agent identity/verification). AWS Bedrock AgentCore Payments shipped May 7, 2026, making x402 hyperscaler-native. The infrastructure layer for agentic commerce is now production. The protocols are layered, not competing in most cases.
• Agentic Memory — How AI Agents Remember Across Sessions Agentic memory is the architecture that lets AI agents retain useful information across sessions, tasks, and time. The category covers everything from simple conversation history to learned skills, persistent project context, and cross-session preference accumulation. Memory is the difference between an agent that re-explains itself every time and one that compounds context over weeks. The discipline of designing memory well is part of agent engineering. Anthropic's Dreaming (May 2026) is the first platform-level realization of automated semantic+procedural memory consolidation.
• Claude Managed Agents — Anthropic's Agent Infrastructure Ready-made infrastructure for running AI agents without building your own orchestration, sandboxes, or tool execution. May 2026 update: Dreaming (between-session memory consolidation), Outcomes GA, multi-agent orchestration GA, Cowork enterprise features, 10 finance + 20 legal templates.
• Prompt Caching — The Production Cost-Optimization Layer for LLM Applications Prompt caching reuses LLM input tokens across requests, cutting input-token costs by up to 90% (Anthropic cache reads are 10% of base price; OpenAI cached inputs run 75-90% cheaper). Combined caching strategies achieve 70-80% total cost reduction in production. The 2026 production landscape: Anthropic cache_control markers with 5-min default TTL (1-hour extended), OpenAI automatic prompt caching, semantic caching via vector similarity (Redis, GPTCache). Distinct from KV caching (model-internal) and agentic memory (cross-session persistence).
• Embeddings — How AI Turns Meaning Into Numbers Embeddings are numerical representations of text (or images, audio, video) that preserve semantic similarity. Two texts about the same topic have similar embedding vectors, even when they use different words. Embeddings are the foundational layer under semantic search, RAG, recommendation systems, and most AI applications that need to match meaning rather than match keywords.
• Tool Use — How AI Agents Reach Out of the Model Tool use is the capability that lets an AI model call external functions — search, calculators, APIs, databases, code execution — rather than generating answers from training alone. Tool use is what turns a chatbot into an agent. It is the technical foundation under all agentic-commerce, agent-engineering, and managed-agents work. The discipline of designing tools well determines whether an agent is reliable or theatrical.
• What's the Break-Even Point for Managed Agents vs. Self-Hosted? Exploring when Anthropic's managed infrastructure becomes more expensive than building your own
• Model Context Protocol (MCP) — Connecting AI to Your Systems Open standard for connecting AI applications to databases, tools, and enterprise systems
• Claude Managed Agents vs. DIY Agent Infrastructure When to use Anthropic's managed platform vs. building your own agent orchestration with Messages API