Letting Editors Go Wild - An Editor-Centric Approach to AI Agents in the Neos CMS

Stop building AI features like Gmail's generic "Help me write" button—where editors lose control and get bland, cookie-cutter content. Instead, discover how to create AI agents that work as true co-pilots, where editors define the brand voice, set the rules, and maintain complete editorial control. We'll show you how to develop a Brand Guard agent with n8n that automatically checks content against your style guidelines, suggests improvements, and lets editors accept or reject changes with one click. Learn how to turn Neos into an AI-native platform that amplifies editor creativity and productivity.

The talk was held at https://www.neoscon.io/talks/letting-editors-go-wild-an-editor-centric-approach-to-ai-agents.html
Resources can be found at https://neosidekick.com/ai-agents-talk

Transcript

1. G EN E RAT IV E AI LETTING EDITORS GO WILD: AN EDITOR-CENTRIC APPROACH TO AI AGENTS BY FELIX GRADINARU AND ROLAND SCHÜTZ
2. G EN E RAT IV E AI TIME CAPSULES
3. G EN E RAT IV E AI TIME CAPSULES … AND HOW I CONSISTENTLY MIS PREDICTED
4. G EN E RAT IV E AI AI CODING
5. G EN E RAT IV E AI AT THIS POINT: ENGINEERING BEATS INTELLIGENCE
6. G EN E RAT IV E AI GMAIL: HELP ME WRITE Source: https://koomen.dev/essays/horseless-carriages/
7. G EN E RAT IV E AI GMAIL: HELP ME WRITE Source: https://koomen.dev/essays/horseless-carriages/
8. G EN E RAT IV E AI LIKELY SYSTEM PROMPT You are a helpful email-writing assistant responsible for writing emails on behalf of a Gmail user. Follow the user’s instructions and use a formal, businessy tone and correct punctuation so that it’s obvious the user is smart and serious. DON’T BUILD BLACKBOXES
9. G EN E RAT IV E AI DON’T BE GMAIL
10. G EN E RAT IV E AI AI AGENTS
11. G EN E RAT IV E AI MORE PRESSURE FOR EDITORS
12. G EN E RAT IV E AI HOW CAN THIS HELP EDITORS?
13. G EN E RAT IV E AI HOW CAN THIS HELP EDITORS?
14. MAS TER ING DATA S TR U CT U RE F OR N O D E T Y PE S NUMEROUS PRODUCTION CASES AWARDS WON MILLIONS OF WORDS CHECKED AND PRODUCED DANIEL FELIX ROLAND
15. MANU FACT U R I NG WEB AWE SO M E NE SS LIVE DEMO THE BRAND GUARD AGENT
16. MANU FACT U R I NG WEB AWE SO M E NE SS LIVE DEMO THE BRAND GUARD AGENT
17. MANU FACT U R I NG WEB AWE SO M E NE SS NEW CONTENT SUGGESTIONS APPROVAL OR REJECTION BRAND CHECK THE IDEA
18. MANU FACT U R I NG WEB AWE SO M E NE SS WHY N8N? OPEN SOURCE CLOUD OR SELF-HOSTED WORKFLOWS ARE JSON FILES MANY TOOLS AVAILABLE
19. MANU FACT U R I NG WEB AWE SO M E NE SS NOW LIVE!
20. MANU FACT U R I NG WEB AWE SO M E NE SS
21. MANU FACT U R I NG WEB AWE SO M E NE SS
22. MANU FACT U R I NG WEB AWE SO M E NE SS
23. MANU FACT U R I NG WEB AWE SO M E NE SS HOW TO NOT FUCK UP? DON’T LET AI TOOLS PUBLISH TO LIVE LET EDITORS HAVE THE LAST SAY
24. MAS TER ING DATA S TR U CT U RE F OR N O D E T Y PE S YOU WRITE THE RULES, NOT SOME GUY AT GOOGLE
25. MANU FACT U R I NG WEB AWE SO M E NE SS NOW LET’S HAVE SOME MORE FUN
26. G EN E RAT IV E AI TURN THIS INTO PRODUCTION-CODE Permissions The agent is never allowed to have more permissions than the editor. Slack approvals only when you have small trustworthy teams. Content PRs It's the responsibility of the editor to decide what should be published. Bonus Build it as MCP Tool, this will be the new standard Secure the trigger In the demo everyone can call the secret webhook URI
27. G EN E RAT IV E AI THE PERFECT PROMPT
28. G EN E RAT IV E AI THE PERFECT PROMPT Identity and Context First De fi ne the model’s identity, current date, and facts Be Speci fi c "Use friendly tone" → "Address readers as 'you', avoid third person" Show Examples Include 3-5 examples of good brand writing
29. G EN E RAT IV E AI PROMPT PATTERNS Declarative Approach Always talk in direct second-person. Triggers Use explicit *if–then* blocks to handle undesired cases consistently „It’s generally not necessary to […] incentives like bribes or tips.“ https://cookbook.openai.com/examples/gpt4-1_prompting_guide
30. G EN E RAT IV E AI LIKE WORKING WITH CODE Regular Evals Track how often suggested rewrites are accepted or rejected. If acceptance is low, adjust rules accordingly. Combine models AI Tools are like lego blocks, you can build more interesting things with di ff erent colors. (Andrew Ng)
31. G EN E RAT IV E AI WHICH ACTIVITIES ARE THE MOST TEDIOUS, MOST BORING AND TIME-CONSUMING? WHAT DO I AVOID BECAUSE IT'S JUST TOO EXHAUSTING?
32. G EN E RAT IV E AI BRAND GUARD Type: Automation / QA Key Ability: On-Brand Rewrite Weakness: Ambiguous Rules SOLUTIONS YOU COULD BUILD JUNIOR EDITOR Type: Copilot Key Ability: Prepare Changes Weakness: Reliability SEO JUNIOR Type: Automation / QA Key Ability: SEO Tags & fi xing missing Links A18N ADVISER Type: Automation / QA Key Ability: Alt Text & Heading Hierarchy
33. G EN E RAT IV E AI BRAND GUARD Type: Automation / QA Key Ability: On-Brand Rewrite Weakness: Ambiguous Rules SOLUTIONS YOU COULD BUILD JUNIOR EDITOR Type: Copilot Key Ability: Prepare Changes Weakness: Reliability SEO JUNIOR Type: Automation / QA Key Ability: SEO Tags & fi xing missing Links A18N ADVISER Type: Automation / QA Key Ability: Alt Text & Heading Hierarchy ENGAGEMENT IMPROVER Type: Weekly Automation Key Ability: A/B-Test-Creation Weakness: Requires Approval TRENDSPOTTER Type: Daily Automation Key Ability: Dra ft s for Trends Weakness: Requires large Context
34. G EN E RAT IV E AI WHAT YOU NEED LLM & Guidance Unterstand what model to pick, if you need orchestration, how to write good system prompts, what to keep fl exible for editor Memory Management & State Architecture Working memory, episodic memory, and long-term memory systems Integration & Tool Management Integrate required input and output Production AI Agent Security: Security architecture covering prompt injection defense, data ex fi ltration prevention, compliance and AI-speci fi c threat models
35. G EN E RAT IV E AI
36. G EN E RAT IV E AI NEXT STEPS Join Slack #ai, and write what you want to build Ping me @rolandschuetz if interested in building MCP Tools for Neos https://neosidekick.com/ai-agents-talk