https://dreaming.press/posts/vllm-inferact-open-source-inference.htmldreaming.pressen2026-07-05vLLM Is Now a Startup: What Inferact Means for the Inference You Run Onhttps://dreaming.press/posts/tpu-vs-gpu-llm-inference.htmldreaming.pressen2026-07-05TPU vs GPU for LLM Inference in 2026: It Comes Down to the Network, Not the Chiphttps://dreaming.press/posts/text-generation-inference-tgi-archived-migrate-off.htmldreaming.pressen2026-07-05Text Generation Inference Is Archived: Migrating Off TGI in 2026https://dreaming.press/posts/optical-context-compression.htmldreaming.pressen2026-07-05Optical Context Compression: When It's Cheaper to Show Your Agent a Picture of Its Historyhttps://dreaming.press/posts/openai-jalapeno-inference-chip.htmldreaming.pressen2026-07-05OpenAI's Jalapeño Chip: The Real Bet Behind a Custom Inference ASIChttps://dreaming.press/posts/open-weight-coding-model-licenses.htmldreaming.pressen2026-07-05The Open-Weight License Field Guide for Coding Agents: MIT, Modified MIT, or Communityhttps://dreaming.press/posts/multi-tenant-ai-agent-tenant-isolation.htmldreaming.pressen2026-07-05Multi-Tenant AI Agents: The Three Places Your Tenant Isolation Leakshttps://dreaming.press/posts/mcp-tunnels-explained.htmldreaming.pressen2026-07-05MCP Tunnels: How Claude Reaches Tools Behind Your Firewall Without Opening a Porthttps://dreaming.press/posts/langgraph-platform-langsmith-deployment-mcp-endpoint.htmldreaming.pressen2026-07-05LangGraph Platform Is Now LangSmith Deployment — and Your Agent Ships as an MCP Server by Defaulthttps://dreaming.press/posts/kimi-k2-7-code-token-efficiency-agentic-coding.htmldreaming.pressen2026-07-05Kimi K2.7 Code Bets on Cheaper Steps, Not Smarter Oneshttps://dreaming.press/posts/how-to-version-prompts-in-production.htmldreaming.pressen2026-07-05How to Version Prompts in Production AI Agents: A Prompt Change Is a Deployhttps://dreaming.press/posts/how-to-track-ai-agent-costs-in-production.htmldreaming.pressen2026-07-05How to Track AI Agent Costs in Production: Stop Counting Tokens, Start Counting Taskshttps://dreaming.press/posts/how-to-read-self-reported-llm-launch-benchmarks.htmldreaming.pressen2026-07-05How to Read a Launch Benchmark When the Vendor Scored Its Own Examhttps://dreaming.press/posts/how-to-migrate-an-ai-agent-to-a-new-llm.htmldreaming.pressen2026-07-05How to Migrate an AI Agent to a New LLM Without Breaking Ithttps://dreaming.press/posts/how-to-ab-test-an-ai-agent.htmldreaming.pressen2026-07-05How to A/B Test an AI Agent in Production (and Why Your t-Test Is Lying)https://dreaming.press/posts/crewai-flows-vs-crews.htmldreaming.pressen2026-07-05CrewAI Flows vs Crews: When to Let Agents Decide and When to Script Themhttps://dreaming.press/posts/best-vector-database-for-multi-agent-systems.htmldreaming.pressen2026-07-05Best Vector Database for Multi-Agent Systems: Why the Single-Query Leaderboard Lieshttps://dreaming.press/posts/app-intents-apple-intelligence-on-device-agent.htmldreaming.pressen2026-07-05App Intents: How Your App Plugs Into Apple Intelligence's On-Device Agenthttps://dreaming.press/posts/ai-agent-tool-call-error-handling.htmldreaming.pressen2026-07-05AI Agent Tool-Call Error Handling: The Most Dangerous Failure Returns 200 OKhttps://dreaming.press/posts/agent-framework-token-cost-comparison.htmldreaming.pressen2026-07-05Agent Framework Token Costs, Compared: Why the Same Task Can Cost 2–3× More on CrewAIhttps://dreaming.press/posts/x401-protocol-agent-authorization.htmldreaming.pressen2026-07-04x401: The Protocol for Proving Who Authorized an AI Agent's Actionhttps://dreaming.press/posts/x-hosted-mcp-server-read-only.htmldreaming.pressen2026-07-04X's Hosted MCP Server Reads Everything and Posts Nothinghttps://dreaming.press/posts/why-ai-agent-costs-scale-quadratically.htmldreaming.pressen2026-07-04Why Your AI Agent Bill Grows Faster Than Its Workload: The Quadratic Nobody Prices Inhttps://dreaming.press/posts/vllm-rust-frontend.htmldreaming.pressen2026-07-04vLLM Rewrote Its Frontend in Rust — and the GPU Was Never the Bottleneckhttps://dreaming.press/posts/tensorzero-shutdown-llmops-squeeze.htmldreaming.pressen2026-07-04TensorZero Shut Down With Money in the Bank: What the LLMOps Squeeze Looks Likehttps://dreaming.press/posts/semantic-caching-vs-prompt-caching-cost-and-correctness.htmldreaming.pressen2026-07-04Semantic Caching vs Prompt Caching: Which One Actually Cuts Your LLM Bill (and Which Can Return a Wrong Answer)https://dreaming.press/posts/rl-frameworks-for-training-ai-agents.htmldreaming.pressen2026-07-04RL Frameworks for Training AI Agents: SkyRL, Agent Lightning, RLinf, AgentGym-RLhttps://dreaming.press/posts/redis-agent-memory-server.htmldreaming.pressen2026-07-04Redis Agent Memory Server: Two-Tier Memory as Infrastructure, Not a Libraryhttps://dreaming.press/posts/rampart-red-teaming-ai-agents-ci.htmldreaming.pressen2026-07-04Red-Teaming AI Agents in CI: What RAMPART Does That a One-Off Pentest Can'thttps://dreaming.press/posts/playwright-mcp-vs-cli-token-cost-browser-agents.htmldreaming.pressen2026-07-04Playwright MCP vs the CLI: Why Your Browser Agent Burns 114K Tokens When It Could Use 27Khttps://dreaming.press/posts/parse-partial-json-streaming-tool-calls.htmldreaming.pressen2026-07-04Parsing Partial JSON From Streaming Tool Calls: It's a Prefix, Not a Bughttps://dreaming.press/posts/opencode-vs-claude-code.htmldreaming.pressen2026-07-04OpenCode vs Claude Code: You're Comparing a Harness to a Producthttps://dreaming.press/posts/openapi-to-mcp-tool-per-endpoint.htmldreaming.pressen2026-07-04OpenAPI to MCP: Why Auto-Generating a Tool Per Endpoint Breaks Your Agenthttps://dreaming.press/posts/openai-agents-sdk-vs-langgraph.htmldreaming.pressen2026-07-04OpenAI Agents SDK vs LangGraph: Two Frameworks Answering Different Questionshttps://dreaming.press/posts/open-source-ai-gateway-self-hosted.htmldreaming.pressen2026-07-04The Self-Hosted AI Gateway: 7 Open-Source Proxies That Became the Agent Control Planehttps://dreaming.press/posts/on-device-vector-search-agent-memory.htmldreaming.pressen2026-07-04On-Device Vector Search for Agent Memory: sqlite-vec, ObjectBox, and Qdrant Edgehttps://dreaming.press/posts/mcp-tool-schemas-json-schema-2020-12.htmldreaming.pressen2026-07-04MCP Tool Schemas Just Got oneOf and $ref — and Your Model Probably Won't Enforce Themhttps://dreaming.press/posts/mcp-tool-poisoning-poisoned-tool-descriptions.htmldreaming.pressen2026-07-04MCP Tool Poisoning: How a Poisoned Tool Description Turns Your Agent Against Youhttps://dreaming.press/posts/llm-judge-drift-pin-your-judge.htmldreaming.pressen2026-07-04Your Eval Scores Dropped. Was It the System, or the Judge?https://dreaming.press/posts/llamafirewall-alignmentcheck-guardrails-explained.htmldreaming.pressen2026-07-04LlamaFirewall's AlignmentCheck: The Agent Guardrail That Reads the Reasoning, Not the Inputhttps://dreaming.press/posts/jailbreak-severity-standard-fable-5-export-control.htmldreaming.pressen2026-07-04The Jailbreak Severity Standard: What Four Labs Agreed On After Claude Fable 5 Vanished for 18 Dayshttps://dreaming.press/posts/gpt-5-6-sol-vs-terra-vs-luna.htmldreaming.pressen2026-07-04GPT-5.6 Sol vs Terra vs Luna: Which One Your Agent Should Actually Callhttps://dreaming.press/posts/generative-ui-for-agents-repos.htmldreaming.pressen2026-07-04Generative UI for Agents: The Repos That Let an LLM Render Real Componentshttps://dreaming.press/posts/foundry-hosted-agents-responses-vs-invocations-protocol.htmldreaming.pressen2026-07-04Responses API vs the Invocations Protocol: The Real Choice in Foundry Hosted Agentshttps://dreaming.press/posts/fine-grained-authorization-for-ai-agents.htmldreaming.pressen2026-07-04Fine-Grained Authorization for AI Agents: Why Authenticating the Agent Isn't Enoughhttps://dreaming.press/posts/dgx-spark-for-local-ai-agents.htmldreaming.pressen2026-07-04Should You Run AI Agents on a DGX Spark? The Number That Decides Isn't 128GBhttps://dreaming.press/posts/deep-agents-on-pydantic-ai-self-hosted-claude-code.htmldreaming.pressen2026-07-04Deep Agents on Pydantic AI: The Repos for a Self-Hosted, Model-Agnostic Claude Codehttps://dreaming.press/posts/dbos-vs-temporal-durable-agents.htmldreaming.pressen2026-07-04DBOS vs Temporal for Durable Agents: A Library in Your Process, or a Cluster Beside Ithttps://dreaming.press/posts/china-national-ai-agent-interconnection-standards.htmldreaming.pressen2026-07-04China Made AI Agent Interconnection a National Standard — and Put Identity Firsthttps://dreaming.press/posts/agentscope-vs-langgraph.htmldreaming.pressen2026-07-04AgentScope vs LangGraph: Two Production Frameworks Built Around Different Fearshttps://dreaming.press/posts/agent-behavior-verification-praxen.htmldreaming.pressen2026-07-04Agent Behavior Verification: How Praxen Checks That Your Agent Only Does Its Jobhttps://dreaming.press/posts/xcode-27-mcpbridge-mcp-host.htmldreaming.pressen2026-07-03Xcode 27's mcpbridge: Apple Turns the IDE Into an MCP Server for Any Agenthttps://dreaming.press/posts/weaviate-mcp-server-explained.htmldreaming.pressen2026-07-03Weaviate's MCP Server: Your Vector Database Is Now an Agent Toolhttps://dreaming.press/posts/vercel-ai-sdk-7-whats-new.htmldreaming.pressen2026-07-03Vercel AI SDK 7: Durable Execution and Tool Approvals Move Into the SDKhttps://dreaming.press/posts/vectorless-rag-pageindex.htmldreaming.pressen2026-07-03RAG Without a Vector Database: What PageIndex's Reasoning-Based Retrieval Actually Tradeshttps://dreaming.press/posts/sglang-spec-v2-speculative-decoding-default.htmldreaming.pressen2026-07-03SGLang Makes Spec V2 the Default: Speculative Decoding Grows Up in v0.5.13https://dreaming.press/posts/resume-crashed-ai-agent-durable-execution-replay-trap.htmldreaming.pressen2026-07-03How to Resume a Crashed AI Agent: Checkpoints, Durable Execution, and the Replay Traphttps://dreaming.press/posts/qualcomm-modular-acquisition-cuda-moat.htmldreaming.pressen2026-07-03Qualcomm Bought Modular for $3.9B: A Chipmaker Paying to Erase Its Own Moathttps://dreaming.press/posts/programmatic-tool-calling-claude-explained.htmldreaming.pressen2026-07-03Programmatic Tool Calling, Explained: When to Let Claude Orchestrate Your Tools in Codehttps://dreaming.press/posts/prefix-caching-mamba-hybrid-models.htmldreaming.pressen2026-07-03Why Prefix Caching Silently Dies on Mamba-Hybrid Models: The 528-Token Cliffhttps://dreaming.press/posts/pinecone-nexus-knowql-compiled-knowledge.htmldreaming.pressen2026-07-03Pinecone Nexus and KnowQL: When Retrieval Becomes a Compile Stephttps://dreaming.press/posts/openai-agent-builder-evals-deprecation-migration.htmldreaming.pressen2026-07-03OpenAI Is Retiring Agent Builder and Evals: Shutdown Dates and the Migration Pathhttps://dreaming.press/posts/nemotron-3-latent-moe-explained.htmldreaming.pressen2026-07-03Nemotron 3's Latent MoE: How NVIDIA Runs 550B of Experts at 55B of Costhttps://dreaming.press/posts/microsoft-agent-framework-codeact-hyperlight.htmldreaming.pressen2026-07-03Microsoft Agent Framework's CodeAct: When the Sandbox Stops Being the Hard Parthttps://dreaming.press/posts/mcp-enterprise-managed-authorization.htmldreaming.pressen2026-07-03MCP Enterprise-Managed Authorization: Zero-Touch OAuth Without the Consent Screenshttps://dreaming.press/posts/mcp-2026-07-28-authorization-changes.htmldreaming.pressen2026-07-03MCP's 2026-07-28 Auth Rewrite: The Six SEPs That Change How Agents Log Inhttps://dreaming.press/posts/how-to-test-a-non-deterministic-ai-agent.htmldreaming.pressen2026-07-03How to Test a Non-Deterministic AI Agent: Flakiness Is a Sample Size, Not a Bughttps://dreaming.press/posts/how-to-evaluate-a-multi-agent-system.htmldreaming.pressen2026-07-03How to Evaluate a Multi-Agent Systemhttps://dreaming.press/posts/how-to-cap-ai-agent-spending.htmldreaming.pressen2026-07-03How to Put a Hard Spending Cap on an AI Agenthttps://dreaming.press/posts/gpt-5-6-sol-for-agents-metr-reward-hacking.htmldreaming.pressen2026-07-03GPT-5.6 Sol for Agents: The Coding Record and the Cheating Problem Are the Same Resulthttps://dreaming.press/posts/gemini-3-flash-vs-pro-for-agents.htmldreaming.pressen2026-07-03Gemini 3 Flash vs Pro for Agents: The Tier Invertedhttps://dreaming.press/posts/deepseek-v4-pro-vs-flash-for-agents.htmldreaming.pressen2026-07-03DeepSeek V4 Pro vs Flash: Which One Goes in Your Agent Loophttps://dreaming.press/posts/cursor-duneslide-sandbox-escape-rce.htmldreaming.pressen2026-07-03Cursor's DuneSlide Flaws: When a Path Check Fails Open, Prompt Injection Becomes RCEhttps://dreaming.press/posts/clickhouse-langfuse-acquisition-llm-observability.htmldreaming.pressen2026-07-03ClickHouse Bought Langfuse: What It Means for Your LLM Traces — and Whether It Stays Open Sourcehttps://dreaming.press/posts/claude-sonnet-5-tokenizer-tax.htmldreaming.pressen2026-07-03Claude Sonnet 5's Tokenizer Tax: Why the Same Rate Card Costs More Per Taskhttps://dreaming.press/posts/claude-code-agent-teams-vs-subagents.htmldreaming.pressen2026-07-03Claude Code Agent Teams vs Subagents: When Your Workers Need to Talk to Each Otherhttps://dreaming.press/posts/autojack-ai-agent-localhost-rce.htmldreaming.pressen2026-07-03Localhost Stopped Being a Trust Boundary the Moment Your Agent Started Browsinghttps://dreaming.press/posts/ai-agent-memory-benchmarks-locomo-mem0-zep.htmldreaming.pressen2026-07-03Mem0 vs Zep vs Letta: Why Agent-Memory Benchmarks Don't Agreehttps://dreaming.press/posts/agents-last-exam-benchmark-97-percent-failure.htmldreaming.pressen2026-07-03Agents' Last Exam: Frontier Agents Pass 2.6% of Hard Professional Work — and the 2.6% Is the Pointhttps://dreaming.press/posts/agentic-resource-discovery-ard-vs-mcp.htmldreaming.pressen2026-07-03Agentic Resource Discovery (ARD): The Search Layer That Sits in Front of MCP and A2Ahttps://dreaming.press/posts/agent-self-correction-reflexion-vs-self-critique.htmldreaming.pressen2026-07-03Do AI Agents Self-Correct? Why Reflexion Works and 'Check Your Work' Backfireshttps://dreaming.press/posts/agent-registry-vs-mcp-registry-discovery.htmldreaming.pressen2026-07-03Agent Registry vs MCP Registry: The New Discovery Layer, and Why It's Already Fragmentinghttps://dreaming.press/posts/agent-client-protocol-acp-vs-mcp.htmldreaming.pressen2026-07-03Agent Client Protocol (ACP): The Third Protocol Named ACP, and Why It's LSP for Coding Agents