GTM OpsHello OperatorVictor's pick
Make Friends With Your Market
useful booth manning advice
Enterprise AIr/artificial
Anyone else feel like AI security is being figured out in production right now?
ai-security-gapsshadow-ai-adoptionprompt-injection-attacksai-agent-permissionssecurity-frameworks-emerging
“Enterprises are averaging 300+ unsanctioned AI apps, and in many cases AI security isn't even owned by security teams”
Key takeaways
- AI security is being figured out in production with enterprises running 300+ unsanctioned AI apps and most lacking dedicated AI security teams
- Attack patterns mirror early-stage tech adoption: prompt injection, over-permissioned agents, and shadow IT rather than sophisticated exploits
- Traditional security knowledge transfers incompletely - prompt injection ≠ SQL injection, agent permissions ≠ API auth - creating expertise gap despite emerging frameworks (OWASP, MITRE ATLAS, NIST)
- AI is accelerating both sides: making it easier for attackers to find weaknesses while defenders struggle with ownership and skill gaps
- Security frameworks exist (OWASP LLM/Agentic Top 10, MITRE ATLAS, NIST AI RMF) but practical implementation expertise is scarce
Why this matters for operators: Companies deploying AI tools without security governance; gap between AI adoption speed and security maturity
I cover AI×GTM intelligence like this every Wednesday.
Get STEEPWORKS WeeklyMore picks
Human-AI IntersectionTechCrunch AIVictor's pick
Meet Shapes, the app bringing humans and AI into the same group chats
this is just ...cool?
- Shapes is positioning as 'Discord with AI characters' - group chat with AI participants
- Represents emerging category of AI-as-peer vs AI-as-tool
- Insufficient detail to assess product-market fit or GTM relevance
ai-collaboration-toolsconversational-aihuman-ai-interaction
AI Developmentn8n Blog
Human-in-the-Loop vs. Human-on-the-Loop: When To Use Each System
- HITL (human-in-the-loop) requires human approval before AI executes critical actions - synchronous control pattern used for high-stakes decisions, compliance requirements, and low-confidence scenarios
- HOTL (human-on-the-loop) allows AI to execute autonomously while humans review results and adjust parameters - asynchronous pattern for scalable operations with exception-based oversight
- Framework applies across use cases: loan approvals, customer emails, social posts, fraud detection, and compliance workflows - choice depends on risk tolerance, regulatory requirements, and operational scale needs
automation-stacksai-policyhuman-first-sales
This analysis was produced using the STEEPWORKS system — the same agents, skills, and knowledge architecture available in the GrowthOS package.