It is not your agent that forgets your safety policy, it is your summarizer
0% violations while the policy is in context, 30% after one compaction. And the rate tracks the model that summarizes, not the one that acts.
AI Agents
Your CRM changed its model, and your contract has no clause for it
Claudeforce changed the default model on several Salesforce surfaces. Five clauses to check in every contract where a model sits inside the product.
Your agents are ready, but your data platform is not
The same models drop from 91% to under 21% accuracy when textbook schemas are replaced by real warehouses.
What your event logs reveal before you deploy AI agents on your processes
Your event log says what the diagram hides. Five questions to ask it, and what to do about the gaps it leaves open.
Devin, OpenHands, Scion: which execution environment for your agents in 2026?
Proprietary sandbox, open source Docker, or GCP orchestrator: three radically different approaches to isolating and running your agents in production.