denn-gubsky

I'm building the agentic runtime and agents running on this platform. The keystone of agentic work is "allowlists vs denylists" choice. On my platform I use allowlists for everything (tools, MCPs, skills, memories etc.). Agents work in a sandbox and have only allowed resources they need for functioning. Nothing more. The LLM is a probabilistic by nature, it can try to access or delete the resource without a reason, it may follow the prompt injection, it may leak the secret or private data. So all this should be controlled by the agentic runtime. In my system I use specialized agents as team players - each agent does it's part of work using available resources. This way you can keep all needed data in the model's context window without compaction and optimize resources control. Ask more specific question if you need a detailed answer.

I develop 5-10 times more complex projects now without anticipation to never complete them. And I can do it alone with team of agents running 24/7. Maybe these new projects will not be used by anyone, but I feel more confident in previously unknown for me domains like robotics, agentic runtimes or GraphRAG. So my advice: just keep evolving.

Thanks for the useful article. I'm building a hierarchy chunked graph document memory as the project knowledge base in my agentic runtime. After reading your articles I will pay more attention into semantic search and retrieval methods. I also started from linked MD documents, but they are too big and contaminate agent's context, so I moved on to the chunked graph model for more selective retrieval operations.

Recently I upgraded my TrueNAS box with Ryzen 7 8900 APU + 96GB DDR5. This config runs Ollama + loomcycle (my agentic Go runtime) + loomboard (the agentic UI), all self-hosted. I tired several models and Ollama configurations. For me is important to fit the whole model in GPU and run multiple agents+tool in parallel (see the blog for details: https://loomcycle.dev/blog/local-llms-on-truenas-and-the-fro...). Gemma4 is fast, but hallucinating a lot, especially when using tools. My current favorite is qwen3.6:27b. This is smart and reliable model, which can run tools in agentic configuration, do long researches, generate code, formulas and diagrams. On my hardware the qwen3.6 speed is about 9-13 tok/s, which is not high, but acceptable. The most tricky part is context size balancing for multiple model instances in limited memory. Ollama splits the context memory between model instances and may fail to run if agentic runtime allocates larger context than size than it is available in the context storage.

I'm doing a lot of QA in my startups. There are several layers: 1. One model (Opus) writes a code, another model (Sonnet, Qwen, Kimi) writes unit tests using requirements and code; 2. Full code review by me. Just to understand what is going on in the codebase; 2. Integration tests are running with PlayWright MCP. Another model validates UI per requirements; 3. Substantive runtime tests prepared by me or human QA team. All features should be included into testing, plus regression testing of existing features, plus edge cases. QA in AI era becomes more important than coding skills. So keep it in a good shape.

In my experience, AI solves the implementation part well - if you know explicitly what should be implemented. If your engineering understanding is fuzzy, the implementation will be fuzzy too.

Engineering and architectural skills matter more now than before. Previously, you had time for rethinking and rearchitecting while coding. That timeline is collapsing - your ideas become real in hours, the bad ones included.

Verification matters more than implementation. When you ship 20 PRs to verify in a single day, every day, all month, how can you be sure the ideas are right, and the implementation is correct? You need ways to challenge your own system: integration tests, runtime tests, and load tests that make you confident in what you're building.

Three skills become non-negotiable in this AI era (until tomorrow, when the answer probably changes again):

1. System Architecture. Build correct framing for your ideas with room for future extensions. Requirements should be complete, precise, and extensible. Project documentation grows with the code and stays current.

2. Organizing AI agentic teams into verifiable flows. Self-evolving specialized agents that verify each other's output until you're confident the code matches the requirements you actually had.

3. Verification, verification, verification. Integration tests, runtime tests, load tests, experiments. The code an AI produces has to be confirmed correct across every dimension that matters.

Thanks for the article. It's super interesting for me, but I have not found a solution for agents running in runtimes without browser. For example on VPS. I'm using WebFetch/WebSearch now, but in some cases I need authorization and cookies. I'm wondering if there is GUI-less browser available, just for agentic usage?

Is it better than A2A?

I integrated the gbash and benchmarked it. gbash is not production grade yet. There are issues with tools implementation and compatibility (I added findings as issues) and performance issues (avg ~30% slower than system C utilities, grep is way slower). Here is full report: https://loomcycle.dev/blog/bashbox-in-process-shell-sandbox

Thank you for good advise. The agentic runtime I'm developing uses system Bash and has to spawn system processes. The pure Go is better match. Checking the repo.

Interesting library. I think I could use for agents running in virtual volumes. I maintain an agentic runtime written on Go. What is the best way to incorporate bashkit.sh?

I'm curious what to do if agent self-improvement goes a wrong way. How do you evaluate the agent quality? Which way do you return to the last good branch? How to automate this process?

Nice and very fast indexing tool. Installed by Claude code, and indexed by 3.5K nodes codebase in less than 2 seconds. Outstanding. The graph looks good, but navigating with the trackpad is not easy. Would be more convenient if zooming centered on mouse position, not in the center of view. So user can select a node by mouse and zoom directly. Anyways I installed the MCP and will see if code review will work faster now.

Thanks a lot. He is 12. I got books and will try to propose for him. But your comic-like website works better :)

Contacted you on LinkedIn.

This is what was for looking for. I need a visual surface to plan agentic feature development workflow. Don't you mind if I make Paca integration with the AI agentic Go runtime I'm developing? your OpenHands concept with container per agent seem little heavy. I will start from WASM plugin, but in perspective, I would like to make a port replacing the OpenHands on services/ai-agent level.

Thank you for sharing this. I'm going to implement this judgement loop in my AI agenting runtime. Worth to try on local LLM with cloud high-end judge model in a loop + QA model to build and run tests as the second judge. I think both judge and QA agents should have an access to the initial RFC requirements. Seems to be a good approach to save API tokens iterating local LLM as code-writer.

I had similar feeling a couple months ago, but I changed the approach and feel more confident now. First of all, any feature now starts from RFC, then RFC brakes into manageable parts, I can review and understand, by planning. When feature is implemented, I run code review and QA pass. This is mandatory. Then I review PR, even eye-balling makes you more confident. If I have doubts, I write the review comment or ask the model to explain what the fragment does and how it relates to other parts. Then integration runtime testing - I use specially trained QA agent. Them manual testing.

Interesting project. I started from the similar layer and bumped into system load barrier. Did you notice the system load when running 5-10-100 agents in parallel? Does you hardware survive 100 Claude.exe (it's claude.exe on macos too!) instances?

Thanks for your work preparing these materials. Sending the link to my son. Philosophy is not included in his curriculum at all.

Thank you. Gemma4:e4b with temperature 0.9 works pretty good with tools (Bash, Glob, Red/Write and WebFetch) And it is fast. Will try the Qwen 3.6 later today.

Software development always was a way to explain the real task to computer. As developer you should understand the domain, you should understand the solution, you should understand tools and approaches. Coding languages and tools are changing and evolving. I started from IBM360, 8080 and MCS48 assemblers and Fortran. A year ago I used C++, Dart, Go and Python. More than year I do not manually write code at all. But I still a software developer. Tools changing, but software development is the same: applying domain knowledge and business requirements to the computing domain. So yes, coding is solved. But software development still alive.

Interesting project. I'm developing a similar project of wider scale. My project is open-source, so we could exchange ideas and technologies.

Interesting project. I'm developing the opensource agentic runtime (Go). Do you plan the gRPC interface to your library for seamless integration with Go? I already have integrations with mem0 and mem9 (planned) memory backends. Would be nice to integrate your really fast and secure OSS memory backend.

Nice library and fast Go code, which is nice to know. I'm solving the similar problem, but on different level. The agentic runtime (Go too) I'm developing, applying safety guards on different layers, including redaction and compression plugins. Do you only block the agentic request or you can redact it in the middle? How do you solve LLM caching problems? What your agent see when it's request is blocked? Does it try to reformulate the request or just stops?

Thank you for sharing. I'm running Gemma4:12b and qwen-coder-next MoE on my local RTX5080 (yes, 16Gb). Using them for local agentic tasks with tools and MCP tools usage. Both hallucinating on my setup. Do you have a rating for most appropriate local models for local agenting tasks? I also tried kimi2.6, qwen3.6 - no luck.

Sorry, this is wrong page. Comment was to the Claude AI reinforcement learning models training article. Wrong click. sorry again.

try latest gemma4:12b. It fits into 16Gb with 256K context window

Try qwen3-coder or qwen3-coder-next models which fit into your configuration. This is team-of-experts model which may load only actual experts into GPU.

Thanks for sharing. I'm developing agentic tools and interested in mem9 intergration. Reading docs and considering.