The Flipped Interaction Pattern: Agent Asks First
Most agents execute on whatever you give them. This one asks questions first. The Flipped Interaction Pattern: an agent that asks before it acts.
Agentic AI
Research notes on AI governance, agentic architecture, and multi-agent system design, with attention to security, risk, and responsible deployment.
TaskWeaver Stateful Data Analyst
A LiteLLM rebuild of TaskWeaver showing planner-to-executor handoff, stateful data memory, and safe pandas actions.
Aegis Constraint Agent
A small LiteLLM rebuild showing deterministic constraint gates around model-assisted mission review.
Knowledge Gap Discovery Agent
A LiteLLM-native agent that finds product knowledge gaps, ranks evidence, and gates model-written hypotheses.
Smolagents Code Agent Rebuild
A LiteLLM-native rebuild of the smolagents code-agent pattern with approved tool calls, AST validation, and provider tests.
Bayesian Operations Gateway
A LiteLLM-native operations gateway with Bayesian-style risk scoring, escalation thresholds, and an audit trail.
Three hidden patterns that inflate your AI API bill
Most AI API overspending comes from three invisible patterns. This post breaks them down with real examples anyone can follow.
MetaGPT Micro Company Agent
A small LiteLLM-native rebuild showing how roles, actions, messages, and a shared environment create an inspectable multi-agent workflow.
EvoAgentX Core Agent Rebuild
A LiteLLM-native rebuild of an EvoAgentX-style agent loop with one approved tool and clear execution traces.
General Problem Solver Agent
A LiteLLM agent that verifies claims first, then escalates hard mismatches into structured reasoning.