Insights and analysis on enterprise AI
Coverage
Ticket routing, agents, fraud, guardrails, RAG: 7 use cases where Jev, TypeSafe's System One model, decides instead of the LLM.
Deep Agents in an isolated sandbox, partitioned subagents, 40x cheaper reporting: the architecture LangChain documents on its own marketing agent.
Distill your code repositories into verified agentic skills: the method (BAAI) that improves an AI agent by +134% without touching the model.
Your AI agent fails intermittently and nobody knows why. Latency and error rates don't catch silent failures: the answer is in your traces. Discover the AgentOps loop that turns production traces into regression tests and quality gates.
Your AI agents fail silently: infinite loops, unverified outputs, runaway reasoning costs. Harness engineering fixes these failure modes without changing the model. Framework, proven patterns and an audit checklist included.
We ran 8 experiments compressing GPT-2 Large with activation-aware SVD and LoRA recovery. The result: 1.6× inference speedup while surpassing the uncompressed baseline by 13%. Here is what we found and why it matters for enterprise AI deployments.
As AI assistants take hold in the enterprise, two extension patterns dominate modern architectures: Skills and MCP servers. They seem similar, but serve very different purposes. Here's how to choose.
These two terms are everywhere in AI discussions, often used interchangeably. Yet they describe very different realities — and picking the wrong approach for your use case can be costly.
AI analysis, field experience and insights delivered to your inbox. No spam — one publication per month.
Your data is never shared. Unsubscribe in one click.