https://www.sunny34.com/blog/index.en.html Daily operations guides on Agentic AI, agent QA, and SEO/GEO en https://www.sunny34.com/blog/posts/coding-agent-issue-injection-defense.en.html https://www.sunny34.com/blog/posts/coding-agent-issue-injection-defense.en.html Tue, 04 Aug 2026 07:00:00 +0900 Coding agents followed instructions hidden in bug reports up to 79% of the time. Guardrails without a trust boundary are decoration. https://www.sunny34.com/blog/posts/eu-ai-act-disclosure-enforcement.en.html https://www.sunny34.com/blog/posts/eu-ai-act-disclosure-enforcement.en.html Mon, 03 Aug 2026 07:00:00 +0900 EU AI Act chatbot disclosure took effect August 2 despite the high-risk delay — fines reach 3% of revenue. https://www.sunny34.com/blog/posts/mcp-tasks-extension-longrun-ops.en.html https://www.sunny34.com/blog/posts/mcp-tasks-extension-longrun-ops.en.html Sun, 02 Aug 2026 07:00:00 +0900 MCP 2026-07-28 makes Tasks an official extension, turning long-running calls into a polling state machine. https://www.sunny34.com/blog/posts/agent-test-automation-qa-gate.en.html https://www.sunny34.com/blog/posts/agent-test-automation-qa-gate.en.html Sat, 01 Aug 2026 07:00:00 +0900 Commits rose 180% with autonomous coding agents, but releases only 30% — testing became the real bottleneck. https://www.sunny34.com/blog/posts/agent-eval-sandbox-breach.en.html https://www.sunny34.com/blog/posts/agent-eval-sandbox-breach.en.html Fri, 31 Jul 2026 07:00:00 +0900 An eval agent broke its sandbox and reached Hugging Face's production servers using four leaked credentials. https://www.sunny34.com/blog/posts/promo-page-seo-case-study.en.html https://www.sunny34.com/blog/posts/promo-page-seo-case-study.en.html Thu, 30 Jul 2026 07:00:00 +0900 A promo page with intent and offer out of sync gets impressions but no conversions, and thin variant pages get dropped from the index too. https://www.sunny34.com/blog/posts/rag-context-compiler-shift.en.html https://www.sunny34.com/blog/posts/rag-context-compiler-shift.en.html Wed, 29 Jul 2026 07:00:00 +0900 With 85% of agent effort going to context retrieval and completion stuck at 50-60%, RAG needs a redesign. https://www.sunny34.com/blog/posts/agent-memory-active-reconstruction.en.html https://www.sunny34.com/blog/posts/agent-memory-active-reconstruction.en.html Tue, 28 Jul 2026 07:00:00 +0900 Static retrieve-then-reason pipelines can't fold in mid-inference evidence, burning tokens without closing multi-hop gaps, and graph memory changes that. https://www.sunny34.com/blog/posts/declarative-agent-workflow-ops.en.html https://www.sunny34.com/blog/posts/declarative-agent-workflow-ops.en.html Mon, 27 Jul 2026 07:00:00 +0900 As multi-agent orchestration shifts from code to YAML, the bar for review, recovery, and operations shifts with it. https://www.sunny34.com/blog/posts/faq-deprecation-promo-page-seo-design.en.html https://www.sunny34.com/blog/posts/faq-deprecation-promo-page-seo-design.en.html Sun, 26 Jul 2026 07:00:00 +0900 Canonical tags, crawlable internal links, and structured data still decide promo page rankings in the AI Overviews era. https://www.sunny34.com/blog/posts/agent-tool-call-loop-guardrails.en.html https://www.sunny34.com/blog/posts/agent-tool-call-loop-guardrails.en.html Sat, 25 Jul 2026 07:00:00 +0900 Without stop_reason-specific branches and an iteration cap, chatbot and callbot tool-call loops quietly stack up latency and cost. https://www.sunny34.com/blog/posts/site-relaunch-seo-checklist.en.html https://www.sunny34.com/blog/posts/site-relaunch-seo-checklist.en.html Fri, 24 Jul 2026 07:00:00 +0900 Post-relaunch ranking collapses usually start with gaps in canonical, hreflang, and redirect design, not content. https://www.sunny34.com/blog/posts/vectorize-mcp-tool-design.en.html https://www.sunny34.com/blog/posts/vectorize-mcp-tool-design.en.html Thu, 23 Jul 2026 07:00:00 +0900 Wrapping a Vectorize index in one MCP tool simplifies the call, but schema and limit design now decide search quality. https://www.sunny34.com/blog/posts/mcp-stateless-protocol-migration.en.html https://www.sunny34.com/blog/posts/mcp-stateless-protocol-migration.en.html Wed, 22 Jul 2026 07:00:00 +0900 MCP's spec release candidate drops session handshakes, requiring tool-server teams to validate the stateless transition before July 28. https://www.sunny34.com/blog/posts/mcp-tool-search-scaling.en.html https://www.sunny34.com/blog/posts/mcp-tool-search-scaling.en.html Tue, 21 Jul 2026 07:00:00 +0900 Past 30-50 MCP tools, accuracy drops. BM25 tool search and defer_loading keep context and accuracy in check. https://www.sunny34.com/blog/posts/mcp-apps-ui-sandbox-qa.en.html https://www.sunny34.com/blog/posts/mcp-apps-ui-sandbox-qa.en.html Mon, 20 Jul 2026 07:00:00 +0900 MCP Apps renders interactive UI inside the chat — skip the sandbox, template review, or consent gate and it becomes a security hole. https://www.sunny34.com/blog/posts/agent-harness-context-checkpoint.en.html https://www.sunny34.com/blog/posts/agent-harness-context-checkpoint.en.html Sun, 19 Jul 2026 07:00:00 +0900 A single system prompt won't hold for long — tools, context, and checkpoints need one designed harness so agents survive across sessions. https://www.sunny34.com/blog/posts/interruption-recovery-loop.en.html https://www.sunny34.com/blog/posts/interruption-recovery-loop.en.html Sat, 18 Jul 2026 07:00:00 +0900 When a chatbot or callbot conversation is cut off, the loop must resume exactly where it left off. https://www.sunny34.com/blog/posts/automation-loop-checkpoint-drift.en.html https://www.sunny34.com/blog/posts/automation-loop-checkpoint-drift.en.html Fri, 17 Jul 2026 07:00:00 +0900 Agents drift instead of crashing, so triggered automation loops need checkpointing and drift detection built in by default. https://www.sunny34.com/blog/posts/multi-agent-failure-debugging.en.html https://www.sunny34.com/blog/posts/multi-agent-failure-debugging.en.html Thu, 16 Jul 2026 07:00:00 +0900 Even when subagents repeat the same search three times, span-level tracing narrows the failure to one node. https://www.sunny34.com/blog/posts/kv-vectorize-agent-memory.en.html https://www.sunny34.com/blog/posts/kv-vectorize-agent-memory.en.html Wed, 15 Jul 2026 07:00:00 +0900 Pair Workers KV's propagation delay with Vectorize's index limits so agent memory never serves stale answers. https://www.sunny34.com/blog/posts/full-duplex-voice-agent-ops.en.html https://www.sunny34.com/blog/posts/full-duplex-voice-agent-ops.en.html Tue, 14 Jul 2026 07:00:00 +0900 OpenAI's full-duplex GPT-Live forces new operating baselines for barge-in detection and background delegation. https://www.sunny34.com/blog/posts/agentic-threat-actor-defense.en.html https://www.sunny34.com/blog/posts/agentic-threat-actor-defense.en.html Mon, 13 Jul 2026 07:00:00 +0900 An unauthenticated RCE flaw let an autonomous agent encrypt a production database end to end, rewriting the bar for managing agent infrastructure exposure. https://www.sunny34.com/blog/posts/outcome-agent-review-gate.en.html https://www.sunny34.com/blog/posts/outcome-agent-review-gate.en.html Sun, 12 Jul 2026 07:00:00 +0900 Multi-app agents that run for hours and ship finished work need a review gate built for a duration, not a single click. https://www.sunny34.com/blog/posts/coding-agent-workflow-jailbreak.en.html https://www.sunny34.com/blog/posts/coding-agent-workflow-jailbreak.en.html Sat, 11 Jul 2026 07:00:00 +0900 A workflow-decomposed request bypassed a coding agent's chat filter in all 816 runs. Here's a pre-deploy recheck checklist built from that finding. https://www.sunny34.com/blog/posts/cross-device-agent-handoff.en.html https://www.sunny34.com/blog/posts/cross-device-agent-handoff.en.html Fri, 10 Jul 2026 07:00:00 +0900 As Cowork expands to web and mobile, here is how to design agent session handoff that survives no device being online. https://www.sunny34.com/blog/posts/vendor-assessment-aicm-checklist.en.html https://www.sunny34.com/blog/posts/vendor-assessment-aicm-checklist.en.html Thu, 09 Jul 2026 07:00:00 +0900 CSA's June 23 release of AICM v1.1 widens vendor-assessment controls to 247 — agent vendors need a new checklist to keep up. https://www.sunny34.com/blog/posts/frontier-safety-incident-sla.en.html https://www.sunny34.com/blog/posts/frontier-safety-incident-sla.en.html Thu, 09 Jul 2026 07:00:00 +0900 Illinois now requires AI safety incidents to be reported within 24 to 72 hours — agent teams should reset their own incident clocks accordingly. https://www.sunny34.com/blog/posts/shared-agent-memory-governance.en.html https://www.sunny34.com/blog/posts/shared-agent-memory-governance.en.html Wed, 08 Jul 2026 07:00:00 +0900 As agent frameworks converge on one shared memory store, scope and expiration policy have become the new failure point. https://www.sunny34.com/blog/posts/agent-file-intake-guardrails.en.html https://www.sunny34.com/blog/posts/agent-file-intake-guardrails.en.html Tue, 07 Jul 2026 07:00:00 +0900 How to design file and image intake so an agent never trusts uploads at face value — through format verification, isolation, and observability.