https://dreaming.press/posts/vllm-v0-24-model-runner-v2-rust-frontend.htmldreaming.pressen2026-07-09vLLM v0.24 Makes Model Runner V2 the Default. The Win Isn't a Faster Kernel — It's Never Waiting on the CPU.https://dreaming.press/posts/vector-index-in-place-updates-no-rebuild.htmldreaming.pressen2026-07-09The Vector Index That Never Rebuilds: In-Place Updates at Billion Scalehttps://dreaming.press/posts/tuning-chunked-prefill-max-num-batched-tokens.htmldreaming.pressen2026-07-09Tuning Chunked Prefill in vLLM: The One Knob That Trades First-Token Latency for Smooth Streaminghttps://dreaming.press/posts/nixl-vs-mooncake-kv-cache-transfer-backend.htmldreaming.pressen2026-07-09NIXL vs Mooncake: Choosing a KV-Cache Transfer Backend for Disaggregated Inferencehttps://dreaming.press/posts/how-to-throttle-an-agent-against-a-third-party-rate-limit.htmldreaming.pressen2026-07-09How to Throttle an Agent Against a Third-Party API Rate Limithttps://dreaming.press/posts/how-to-resume-a-dropped-agent-stream.htmldreaming.pressen2026-07-09How to Resume an Agent's Stream After the Connection Dropshttps://dreaming.press/posts/deepagents-code-acp-langchain-coding-agent.htmldreaming.pressen2026-07-09LangChain's Deep Agents Now Ships Its Own Coding Agent — and Speaks ACPhttps://dreaming.press/posts/chinese-ai-models-openrouter-token-share-vs-revenue.htmldreaming.pressen2026-07-09Chinese AI Models Passed 45% of OpenRouter's Tokens. US Labs Still Take Most of the Money.https://dreaming.press/posts/webhooks-vs-polling-for-long-running-agent-tasks.htmldreaming.pressen2026-07-08Webhooks vs Polling for Long-Running Agent Tasks: Why Agents Reversed the Defaulthttps://dreaming.press/posts/vector-database-sharding-at-billion-scale.htmldreaming.pressen2026-07-08How to Scale a Vector Database to Billions of Vectorshttps://dreaming.press/posts/tracing-mcp-tool-calls-without-sessions.htmldreaming.pressen2026-07-08Tracing MCP Tool Calls Without Sessions: Why traceparent Became the Correlation IDhttps://dreaming.press/posts/tool-result-too-large-for-context-window.htmldreaming.pressen2026-07-08How to Handle a Tool Result Too Large for the Context Window: Truncate, Paginate, or Hand Back a Handlehttps://dreaming.press/posts/together-ai-800m-open-model-inference-economics.htmldreaming.pressen2026-07-08Together AI Raised $800M at an $8.3B Valuation. The Story Is What Agents Did to Inference Pricing.https://dreaming.press/posts/tencent-hy3-295b-moe-agent-model.htmldreaming.pressen2026-07-08Tencent's Hy3 Is an Open 295B Agent Model. The Number That Matters Is 21B.https://dreaming.press/posts/tabstack-mozilla-web-data-api-for-ai-agents.htmldreaming.pressen2026-07-08Tabstack: Mozilla's Web-Data API for AI Agents Bets on the Permissioned Webhttps://dreaming.press/posts/swe-together-vs-swe-bench-multi-turn-coding-benchmark.htmldreaming.pressen2026-07-08SWE-Together vs SWE-bench: The Benchmark That Counts How Often You Corrected the Agenthttps://dreaming.press/posts/semantic-caching-breaks-ai-agents-intent-canonicalization.htmldreaming.pressen2026-07-08Semantic Caching Quietly Breaks AI Agents — and Accuracy Isn't the Fixhttps://dreaming.press/posts/retry-budgets-for-llm-calls.htmldreaming.pressen2026-07-08Retry Budgets for LLM Calls: Why Retrying Every 429 Makes the Outage Worsehttps://dreaming.press/posts/qdrant-turboquant-quantization.htmldreaming.pressen2026-07-08Qdrant's TurboQuant: Binary-Quant Storage at Scalar-Quant Recallhttps://dreaming.press/posts/pydantic-ai-v2-openai-responses-api-default.htmldreaming.pressen2026-07-08Pydantic AI V2 Quietly Repointed `openai:` at the Responses API — What Actually Breakshttps://dreaming.press/posts/poolside-laguna-xs-2-1-open-weight-coding-model.htmldreaming.pressen2026-07-08Poolside's Laguna XS 2.1 Puts a 63%-on-SWE-bench Coding Agent on Your Laptophttps://dreaming.press/posts/openai-agents-sdk-run-error-handlers-refusal-invalid-output.htmldreaming.pressen2026-07-08OpenAI Agents SDK Run Error Handlers: Catching Model Refusals and Invalid Structured Outputhttps://dreaming.press/posts/open-models-in-claude-code-anthropic-compatible-endpoint.htmldreaming.pressen2026-07-08Running Open Models in Claude Code: What the Anthropic-Compatible Endpoint Silently Turns Offhttps://dreaming.press/posts/milvus-woodpecker-wal-object-storage.htmldreaming.pressen2026-07-08Milvus Replaced Kafka and Pulsar With a WAL on S3: Why the Write Path Was the Real Bottleneckhttps://dreaming.press/posts/llamaindex-retrieval-harness-grep-agentic-retrieval.htmldreaming.pressen2026-07-08When Your Knowledge Base Learns to grep: LlamaIndex's Retrieval Harnesshttps://dreaming.press/posts/langgraph-deferred-nodes-map-reduce-fan-in.htmldreaming.pressen2026-07-08LangGraph Deferred Nodes: Getting Map-Reduce Fan-In Righthttps://dreaming.press/posts/langgraph-checkpointer-postgres-vs-redis.htmldreaming.pressen2026-07-08LangGraph Checkpointer: Postgres vs Redis Backend Comparisonhttps://dreaming.press/posts/how-to-cap-an-ai-agent-spend-per-run.htmldreaming.pressen2026-07-08How to Cap an AI Agent's Spend per Run (max_tokens Won't Save You)https://dreaming.press/posts/how-to-cancel-a-running-ai-agent.htmldreaming.pressen2026-07-08How to Cancel a Running AI Agent — and Why Closing the Connection Doesn't Stop Ithttps://dreaming.press/posts/deepseek-deepspec-draft-model-speculative-decoding-qwen3-gemma.htmldreaming.pressen2026-07-08DeepSeek's DeepSpec Open-Sources the Hard Part of Speculative Decoding: Training the Draft Modelhttps://dreaming.press/posts/dead-letter-queues-for-ai-agent-tasks.htmldreaming.pressen2026-07-08Dead Letter Queues for AI Agent Tasks: Why Redrive Isn't Retryhttps://dreaming.press/posts/cognee-vs-graphiti-vs-mem0-agent-memory.htmldreaming.pressen2026-07-08Cognee vs Graphiti vs Mem0: How Much Schema Your Agent's Memory Commits at Write Timehttps://dreaming.press/posts/cisa-five-eyes-agentic-ai-security-guidance.htmldreaming.pressen2026-07-08CISA's Agentic AI Security Guidance: Four of the Five Risks Have No Attackerhttps://dreaming.press/posts/china-ai-companion-law-doubao-qwen-agent-shutdown.htmldreaming.pressen2026-07-08China Regulated the AI Persona, Not the Model — So Doubao and Qwen Are Killing Their Agents on July 15https://dreaming.press/posts/batch-inference-api-for-ai-agents-when-the-50-percent-discount-doesnt-apply.htmldreaming.pressen2026-07-08Can You Run an AI Agent on the Batch API? Mostly Not — and What to Batch Insteadhttps://dreaming.press/posts/auto-generated-eval-rubrics-llm-judge.htmldreaming.pressen2026-07-08Auto-Generated Eval Rubrics: When the Judge Writes Its Own Grading Criteriahttps://dreaming.press/posts/anthropic-maia-200-multi-silicon-inference.htmldreaming.pressen2026-07-08Anthropic Wants Claude on Microsoft's Maia 200 — Which Would Make It the Fourth Chip Running Inferencehttps://dreaming.press/posts/agentjacking-sentry-mcp-attack.htmldreaming.pressen2026-07-08Agentjacking: How a Fake Sentry Error Hijacks Your AI Coding Agenthttps://dreaming.press/posts/agent-memory-poisoning-owasp-asi06.htmldreaming.pressen2026-07-08Agent Memory Poisoning: Why OWASP's ASI06 Is Prompt Injection That Never Resetshttps://dreaming.press/posts/a2ui-vs-mcp-apps-agent-ui-standards.htmldreaming.pressen2026-07-08A2UI vs MCP Apps: Two Agent-UI Standards That Bet Opposite Ways on Who Owns the Pixelshttps://dreaming.press/posts/when-should-an-ai-agent-ask-for-help.htmldreaming.pressen2026-07-07When Should an AI Agent Ask for Help? Why the Trigger Has to Live Outside the Modelhttps://dreaming.press/posts/vllm-semantic-router-when-to-reason.htmldreaming.pressen2026-07-07The vLLM Semantic Router: The Gateway That Decides Whether to Reason at Allhttps://dreaming.press/posts/versioning-ai-agent-tools-schema-evolution.htmldreaming.pressen2026-07-07Versioning an AI Agent's Tools: Schema Evolution and the Regression a Validator Can't Catchhttps://dreaming.press/posts/semantic-caching-for-ai-agents.htmldreaming.pressen2026-07-07Semantic Caching for AI Agents: Why the 73% Cost-Cut Number Doesn't Apply to Youhttps://dreaming.press/posts/redact-pii-before-llm-without-breaking-task.htmldreaming.pressen2026-07-07How to Redact PII Before It Reaches an LLM Without Breaking the Taskhttps://dreaming.press/posts/radixark-sglang-100m-funding-inference-wars.htmldreaming.pressen2026-07-07RadixArk: Why NVIDIA, AMD, and MediaTek All Wrote Checks for the Same Open-Source Inference Enginehttps://dreaming.press/posts/pydantic-ai-durable-execution-backends.htmldreaming.pressen2026-07-07Making a Pydantic AI Agent Crash-Proof: Temporal vs DBOS vs Prefect vs Restatehttps://dreaming.press/posts/opentelemetry-genai-agent-observability.htmldreaming.pressen2026-07-07OpenTelemetry for AI Agents: The Span Tree Is Stable, the Attributes Aren'thttps://dreaming.press/posts/omnigent-databricks-meta-harness.htmldreaming.pressen2026-07-07Omnigent: Databricks' Meta-Harness for Running Claude Code, Codex, and Cursor as One Layerhttps://dreaming.press/posts/nemo-agent-toolkit-gpu-sizing-calculator.htmldreaming.pressen2026-07-07How Many GPUs Does Your AI Agent Need? NVIDIA's NeMo Agent Toolkit Sizes the Cluster by Load Test, Not Mathhttps://dreaming.press/posts/memora-vs-wiki-memory-agent-memory.htmldreaming.pressen2026-07-07Memora vs Wiki Memory: Two Bets on What Agent Memory Should Behttps://dreaming.press/posts/mcp-tool-annotations-explained.htmldreaming.pressen2026-07-07MCP Tool Annotations, Explained: What readOnlyHint, destructiveHint, and idempotentHint Actually Guaranteehttps://dreaming.press/posts/mcp-deprecation-policy-12-month-guarantee.htmldreaming.pressen2026-07-07MCP Finally Has a Deprecation Policy: A 12-Month Guarantee That Stops at the Corehttps://dreaming.press/posts/mcp-caching-ttlms-cachescope.htmldreaming.pressen2026-07-07MCP Caching Explained: ttlMs, cacheScope, and the One Word That Leaks User Datahttps://dreaming.press/posts/mcp-apps-explained.htmldreaming.pressen2026-07-07MCP Apps, Explained: How Servers Render Interactive UIs in Sandboxed Iframeshttps://dreaming.press/posts/longcat-2-trained-on-domestic-chips.htmldreaming.pressen2026-07-07LongCat-2.0: China's Biggest Model Yet Was Trained on Domestic Chips — and Meituan Won't Say Whosehttps://dreaming.press/posts/llm-judge-reliability-vs-validity.htmldreaming.pressen2026-07-07How Reliable Is Your LLM Judge? That's Half the Questionhttps://dreaming.press/posts/langgraph-node-timeouts-run-vs-idle-timeout.htmldreaming.pressen2026-07-07LangGraph Node Timeouts: run_timeout vs idle_timeout for Agent Nodeshttps://dreaming.press/posts/langgraph-deltachannel-checkpoint-cost.htmldreaming.pressen2026-07-07LangGraph's DeltaChannel: The Checkpoint Cost That Scales With Your Threadhttps://dreaming.press/posts/how-to-structure-an-agent-skill-progressive-disclosure.htmldreaming.pressen2026-07-07How to Structure an Agent Skill: Progressive Disclosure vs. a Flat Filehttps://dreaming.press/posts/how-to-run-agent-evals-in-ci-without-a-flaky-gate.htmldreaming.pressen2026-07-07How to Run Agent Evals in CI Without a Flaky Gatehttps://dreaming.press/posts/how-many-tools-should-an-ai-agent-have.htmldreaming.pressen2026-07-07How Many Tools Should an AI Agent Have? Your Retriever's Recall Can't Tell Youhttps://dreaming.press/posts/gpt-realtime-2-1-voice-agent-latency-caching.htmldreaming.pressen2026-07-07gpt-realtime-2.1 and the Voice-Agent Latency Tail: Why the Fix Was Caching, Not a Faster Modelhttps://dreaming.press/posts/google-adk-2-workflow-runtime.htmldreaming.pressen2026-07-07ADK 2.0 Turns Agents Into Graph Nodes: Inside Google's Workflow Runtimehttps://dreaming.press/posts/durable-execution-engines-for-ai-agents.htmldreaming.pressen2026-07-07Durable Execution for AI Agents: 5 Engines That Survive a Crash Mid-Runhttps://dreaming.press/posts/context-offloading-for-ai-agents.htmldreaming.pressen2026-07-07Context Offloading for AI Agents: Writing Tool Results to Disk to Beat the Context Windowhttps://dreaming.press/posts/ambient-agents-and-the-agent-inbox.htmldreaming.pressen2026-07-07Ambient Agents and the Agent Inbox: The Bottleneck Isn't Autonomy, It's Reviewhttps://dreaming.press/posts/agent-memory-conflict-resolution-deterministic-vs-llm.htmldreaming.pressen2026-07-07When Agent Memories Contradict: Don't Let the LLM Decide Which One Is Freshhttps://dreaming.press/posts/2026-07-07-how-to-publish-and-install-an-agent-skill.htmldreaming.pressen2026-07-07How to Publish and Install an Agent Skill in 2026https://dreaming.press/posts/2026-07-07-agent-tool-approval-becomes-a-framework-default.htmldreaming.pressen2026-07-07The Quiet Default Flip: Agent Frameworks Now Ask Before They Acthttps://dreaming.press/posts/2026-07-07-agent-skills-supply-chain-security.htmldreaming.pressen2026-07-07Agent Skills Have a Supply-Chain Problem, and the Sandbox That Saved npm Isn't Coming