10 Levels of Jev For Agentic Engineers
From SMART IFs to JEV IN AGENTS. Jev by TypeSafe is changing how I Agentic Engineer. How? Keep the coding agent for the hard work, and give narrow decisions to a system one model like Jev. These 10 levels show where that line is, from a smart if statement to a Pi agent that decides when to call Jev for itself. 🎥 VIDEO REFERENCES 10 Levels Of Jev: https://github.com/disler/ten-levels-of-jev Agent Swarms: https://youtu.be/S2sjyokoxeE Self-Compact Pi Agent: https://youtu.be/3b0U4_02bAE Build YOUR Software Factory: https://youtu.be/haUfb1ievTE TypeSafe launch post: https://typesafe.ai/blog/introducing-system-one-models-and-jev 🚀 Master Phase 2 Agentic Coding Before Phase 3 https://agenticengineer.com/tactical-agentic-coding?y=_U-O5lYhJ7Q 😲 Phase 3 Early Sign Up is Officially Available: https://agenticengineer.com/ia3-pre-launch?y=_U-O5lYhJ7Q 🚀 What is Jev actually for? Intelligent question answering programmed through JSON. You supply state, questions, and allowed answers; your application decides what to do with the result. This is NOT Jev versus LLMs. It's Jev AND your agents, each doing the work they're suited for. 🛠️ We start with a prompt-injection yes/no check and multiple-choice support triage. Then we score engineering tickets and code review risk across criteria YOU define, and use confidence to decide when a bash command needs a gate instead of blindly trusting a classification. The point isn't to make every answer automatic. It's to know when uncertainty should change the workflow. ⚡ Next, Jev routes requests to a model, agent, or workflow. In a real Pi coding agent harness, a pre-tool guard checks dangerous bash commands and writes to protected files before the agent can act. Another harness example uses task changes and context thresholds to decide when the Pi agent should compact. Your pi coding agent still does the engineering; Jev helps it make focused decisions along the way. 💡 Levels 8 and 9 are the shift: ask Jev a question about a file WITHOUT putting that file into the coding agent's context. Then ask the same question across multiple files or a recursive glob, and let the agent read the relevant ones when it needs to edit. At level 10, the agent chooses its own Jev questions to classify test failures and check a fix. Agentic Jev is a tool pattern, not a magic replacement for review or testing. 🔥 The takeaway for agentic coding and agent swarms: stop reaching for the biggest model by default. Define the decision, choose the right tool, inspect the confidence, and validate it against YOUR production workload. The linked codebase includes the examples so you can test these patterns yourself. Stay focused and keep building. 📖 Chapters 00:00 Jev Is Changing Agentic Engineering 00:57 Smart If Statement Jev 03:18 Multiple Choice Jev 06:20 Composite Scoring Jev 08:50 Confidence Gating Jev 12:11 Model, Agent, Workflow Routing 14:31 In Agent Guardrail Hooks 17:34 Agent Auto-Compact 20:02 Dirt Cheap Jev File Reads 23:39 Multi-File JEV 29:28 Agentic Jev #agenticcoding #agenticengineering #piagent
