Karpathy’s Agentic Engineering finally has proper DevTools! When an agent stops working, the model is only one possible cause. The problem could be a failed tool, a lost connection, an interface update that never appeared, or something earlier in the conversation. @CopilotKit has rebuilt its open-source Inspector around this problem. It sits inside the application and watches the full interaction between the user, the interface, and the agent. When something fails, the Inspector button turns red, names the failure, and opens the spot where it happened. But even a much harder problem is reproducing that failure. This is because agents do not always follow the same path twice. Running the same prompt again may trigger another tool, produce another response, or start with a different application state. Inspector handles this through saved Threads and an isolated Agent Playground. A developer can open the conversation where the problem occurred, choose "Try from here," and copy everything up to that point into the Playground. The original conversation remains unchanged while another response or tool path is tested. Threads can open inside the live application, while assistant messages can jump back to their matching context in Inspector. Everything runs underneath AG-UI, the protocol that carries messages, tool activity, state changes, and other events between the agent and the interface. Those interactions can also feed CopilotKit Intelligence so that when the same issue or useful behavior appears repeatedly, it can become an evidence-backed Insight and a proposed SKILL(.)md file. Developers can review, edit, and approve the improvement before the agent inherits it. The tool also helps developers find a problem, recreate its surrounding context, test another path, and turn repeated lessons into agent improvements. It is open source and included with CopilotKit development builds. Intelligence setup starts with one prompt. Here are the docs: I am testing this extensively and will cover this in more detail soon with a hands-on demo.

@karpathy and I are back! At @sequoia AI Ascent 2026. And a lot has changed. Last year, he coined “vibe coding”. This year, he’s never felt more behind as a programmer. The big shift: vibe coding raised the floor. Agentic engineering raises the ceiling. We talk about what it means to build seriously in the agent era. Not just moving faster. Building new things, with new tools, while preserving the parts that still require human taste, judgment, and understanding.

Karpathy's LLM compiler idea: raw docs -> LLM -> wiki articles -> agent queries at runtime. Everyone's applying it to external data. But your Claude Code session logs are the same thing. Raw folder already exists. Nobody's compiled it yet...🤔

Andrej Karpathy: "A 1B model can be much more effective than most current LLMs. 90% of current models waste a ton of capacity and money on remembering stuff that doesn't matter."

Andrej Karpathy just gave one of the clearest explanations I’ve heard for how people should be thinking about their Hermes agents. Not one agent. All of them. “Remove yourself as the bottleneck.” That line is much bigger than better prompting. Karpathy is talking about arranging your agents so they can operate with less human intervention, keep work moving longer, and accomplish far more from the relatively small amount of direction you give them. That should be the direction you’re pushing every Hermes agent you build. Your research agent shouldn’t need you to tell it what to search next every five minutes. Your monitoring agent shouldn’t depend on you remembering to check something. Your content agent shouldn’t stop after step one when the rest of the workflow is already known. Your development agent shouldn’t need you sitting there approving every tiny action. That doesn’t mean blindly making everything autonomous. It means deliberately looking at every place where you are still unnecessarily sitting in the loop. Give the agent a clear objective. Give it the tools it needs. Give it the right Skills, memory, files and project context. Define the boundaries. Define what success looks like. Give it a reason to start working, whether that’s a schedule, an event, a recurring task or another agent handing work off to it. Most importantly, define what it can handle itself and what actually requires you. That’s the shift. You’re no longer just building something you can chat with. You’re building an operating workflow that can continue doing useful work after you walk away. Karpathy later puts the goal even more directly: get “more agents running for longer periods of time without your involvement.” That might be one of the best north stars you can use for a serious Hermes setup. Because the number of agents you have means almost nothing by itself. You can have 20 Hermes agents and still personally drive every single action. Or you can have five well-designed agents researching, monitoring, organizing, building and reporting while your attention is somewhere else. So instead of asking: How many Hermes agents can I build? Start asking this about every agent you already have: What am I still doing that this agent could be doing without me? That’s where the leverage starts.

HOW KARPATHY'S PROMPT TURNS OBSIDIAN INTO AN ACTUAL SECOND BRAIN IN 15 MINUTES. 1) Install Obsidian and create a vault. 2) Get Claude Code or Cowork pointed at that folder (it needs real filesystem access, which browser based Claude does not have). 3) Paste Karpathy's LLM wiki gist and tell it to implement the pattern. 4) Claude builds the folder structure, the CLAUDE.md schema, and the index automatically. 5) Drop any source in after that and it gets read, linked, and filed without you touching a single folder. Bookmark this. Follow @cyrilXBT