Intelligence for marketing engineers

Key takeaways

• Langfuse v4.8.0 — per-conversation tool approval plus fixed Anthropic token cost tracking
• n8n v2.35.0 — production hardening for AI agent execution with cleaner response handling
• GLM-5.2 — MIT-licensed long-context model joins self-hosted chatbot options
• ChatGPT integration friction — practitioners still struggle connecting LLM outputs to martech tools

Key takeaways

• OpenAI education plugins — structured workflows for teaching and learning, adaptable for customer education
• MiniMax-M3 GGUF — local multimodal inference for image-text tasks and video understanding
• LFM2.5-VL 1.6B — edge-ready vision-language model for resource-constrained devices
• Google Assistant discontinuation — voice marketing touchpoints may need migration to Gemini
• SEO-PPC data gap — unified search visibility workflows remain unsolved for most teams

Key takeaways

• Parable-Granite 8B ships — local agentic tool-use without cloud dependencies
• Circles hits 22% ARPU lift — telco case study ties AI integration to revenue
• Cursor adds Google Workspace plugins — manage Gmail and Docs from your editor
• PPC defaults can cost revenue — context-agnostic optimization flags a risk

Key takeaways

• GPT-5.6 released — claims better efficiency for agents; single source, verify before routing changes
• MiniCPM5-1B adds tool-calling — test against your smallest model for simple agent tasks
• Agent security post-mortem published — review for hardening production workflows
• Prompt injection risks in marketing AI documented — audit inputs and outputs for brand safety
• Attribution vs incrementality clarified — reconcile both to avoid budget misallocation

Key takeaways

• OpenAI Presence — managed voice and chat agents with enterprise governance controls
• Context warehouses — a proposed infrastructure pattern for agent-friendly data access
• ChatGPT for Small Business — training and automation resources for SMB adoption
• Gemini 3.5 Flash Cyber — lightweight model for autonomous security patching

Key takeaways

• Qwen3-8B dialogue fine-tune — built for chat, but needs eval against your prompts
• NVIDIA Cosmos 3 Edge — 7B model for on-device inference, relevant for local-first data processing
• 27B quantized multimodal build — vision-language tasks on consumer GPUs if you analyze creative assets
• AI search monitoring gap — no mature vendor tools for tracking brand mentions in chat answers

Key takeaways

• OpenAI GPT-Red automates agent vulnerability testing — run it against your marketing chatbots before your next deployment
• Hugging Face VoiceEQ benchmarks voice AI quality — test it if voice metrics matter to your IVR or audio workflows
• Russian-language RAG splitter launched — try it if you chunk Russian support content or knowledge bases
• Apple Silicon runs 26B vision model locally — spike a prototype if you need offline visual analysis

Key takeaways

• Multilingual embedding model for CPU — run semantic search locally without GPU costs
• Multimodal chat models on HuggingFace — test image-text dialogue without cloud APIs
• AI shopping agents need structured data — check your product schema for discoverability

Key takeaways

• Unified LLM eval framework — scattered rubric-based judging techniques get combined into a single open-source tool with ensemble judging and bias mitigation built in.
• Agent learning from experience — new research shows LLM agents can extract transferable heuristics from past task trajectories, improving without retraining.
• Cheap LLM evals + selective human audits — a mathematically sound method for finding optimal system configurations when LLM judges are biased but expensive human review is accurate.
• Agent capability hierarchy — structured breakdown of which gaps (tool use, planning, adaptability) cause agent failures on real workplace tasks, relevant for marketing orchestration pipelines.
• Multi-agent coordination under partial information — benchmark reveals stronger individual reasoning doesn't always mean better team coordination, with implications for multi-agent campaign workflows.