Thoughts by human, co-written by AI

i want to write an engineering article with neat/tricky animation preamble for the 100 PRs of the /Users/goga/Documents/goga/backlog-mcp https://github.com/gkoreli/backlog-mcp

i am 117 PRs, and then at some point i moved to making commits and pushing directly, no more PRs, but those initial 100PRs were instrumental for this project. For example, in one of the weekends i created my own UI framework, because I was facing lots of challenges in the UI at that point, so i had no other choise, i ran a spike with react and didn't like it, and no other UI framework aligned with my vision. I want to write like an engineering article almost like research paper but mostly engineering challenges, ideas, and vision that i have for the future of agentic engineering, my thinking, mental model and so on, you will have to mine this information across PRs, project files, ADRs and so on... You can delegate to sub-agents, use Sonnet for exploration and mining information or any hard labor tasks that you have that don't require critical judgment like you as Fable 5 the most advanced model, you are the brain and main writer.

write the .ts based advanced blog, you can have some cross references with rationale and why they matter relative to the article we are writing. Make sure to write a catchy engineering article. Be mindful about SEO so that people find out about us, about backlog mcp, about nisli/core ui framework and about me.

Also you can start another exploration to figure out what it means to write like Goga and what it means to write catchy articles in present tense. read the my personal creative writing skill thats mutual for both of us .agents/skills/personal-essays/SKILL.md - respect it and the article needs to respect it as well

i think the article is talking too much about the numbers as if its like analytics, the numbers are a consequence the article is about the saga, its like a part 1 of the backlog mcp as a product article series, but from the perspective of 100 PRs, what it means, what does it take to be building in open source as Goga, what does it mean to be so determined to be building from personal pain of not having proper task management tools and so on.

i dont like the narrative, the number of pull requests is not the main concept of the article its just the consequence. I want a catchy title, like simple task manager that turned into a robust context/memory engineering tool for my ai agent - what happened after 100+ PRs. And it starts that we have now this robust product still called backlog-mcp that transcended the mcp part of it, as a result i created UI Framework from because i have lots of experience with UI frameworks, i have been using Angular since the angular 2 beta version came out, we were running beta angular 2 in production and since joining AWS i have moved to React heavily and ihave used and experimented with all the UI frameworks. I built nisli/core ui framework to be agentic friendly, barebones, one of the closest native to web. I built entire backlog-mcp and then i built entire blog out of it, I build nisli/static for static rendering (not the SSR), i built Static Site Generation (SSG) for blog to be SEO friendly and it works perfectly. I built the memory layer into backlog mcp. I made entire backlog to be remote friendly and hosted on cloudfront and decided to slightly pivot, to make the local first to be the north star, because of wanting to have all the data locally. I can fix the remote side of the things with synchronization. I built the entire backlog-mcp as a resilient storage system. It is context engineering and memory engineering tool for agentic first engineering.There are so many novel concepts, for example you can't mutate anything from UI, UI is purely readonly, and it is an explicit part of the vision, i dont want to click buttons to CRUD, all the actions are done by backlog-mcp via agents, never manually. There is an entire core so that agent can use cli instead of mcp, one core many consumers. There is a search engine using Orama and we have context RAG. This is where we stand today. Today I am mostly investing around the memory ergonomics and improving the memory. I created the concept of substrates in the core, so that agent doesn't need to know about all the types ahead of time, its on-demand, because agents love on-demand context exposure instead of blasting with huge information. I want catchy title and catchy evidence based initial paragraph that makes backlog mcp sound really enticing.

here is the repo btw: https://github.com/gkoreli/backlog-mcp you can explore. I think i like this as well, i want it to be catchy, like what is he talking about, is it better to say 100 or 117? After 117 Pull Requests, It Wasn’t a Task Manager Anymore

okay then lets keep it: But the original refined to “117 Pull Requests Later…” is the best. It creates curiosity without immediately explaining the trick.

Lets go ahead rewrite with this kinda catchy story + engineering article narrative.

Somewhere in the middle i want you to show some kinda nice timeline of what happened, you can see through the PRs or commits what was the narrative. Another concept that I am havily using and can coin the term for it is, i am using adr driven engineering. ADRs are different from specs, specs get outdated heavily, but for some reason this small mental shift of having ADRs, threaded with context and with front matter, have stayed with me and ever since i am using it everywhere.

another interesting concept or mental model that i developed myself is that, nowadays building is not a problem any more, the problem is building correct things, or what to build. The mental state that I achieved is that things that I build are now truly birthed from the dire desire to address the burning problem that I am facing at the moment, thats how i started with tasks, then i faced too many tasks and needed search, i couldn't see the situation of the tasks so i had to create backlog mcp viewer, but in the ui i didn't want to waste time manually, so i explicitly only added features to the mcp so that if i wanted to do something first added to the core and exposed via mcp to agent, thats how backlog mcp ui thats readonly got birthed, then i added different substrates because i needed more then tasks, at some point i had hundreds of tasks and backlog mcp viewer UI became complicated and i had to do something about the ui framework otherwise the entire app would refresh and i would lose the state of where i was and what i was doing, because at first i started with vanilla native web components only. Then context engineering, then memory, you can see this throughout the entire timeline of the project, this is the central theme for the backlog mcp, build something that solves the problem start using, uncover new problems, address those, keep using them, mostly never anticipated or over-engineered solutions to un-existing or potential theoretical problems. Thats how I see agentic product engineering and thats the philosophy i am following because otherwise people build so many perfect looking products that nobody even including them use. I am still facing problems right now memory is not great, and i dont know how to solve it, but i definitely know i want to improve ergonomics and organize them because i am accumulating over 200+ memory artifacts. One thing that i thought and failed was to use memory injection on every agent turn, that one is bad. Also i am afraid of having stale memory artifacts or misleading or incorrect ones. OR wasting too much tokens on memory. Or i have fomo missing out for not using mem0 or mempalace or some established libraries, but i feel like they over engineer for my needs, i want to build from my problems, not from someone else's problems, but i can definitely bem ore than welcome to get inspired from their ideas.

what about the prompts.md file? i feel like you should update that too?

i want the article to have a personal touch, for the reader to feel and for me as an author to bleed, for example, the fact that i am facing lots of problems and i dont know how to solve them, i have ideas, but they are quite complicated. For example, Obsidian got a lot of traction recently and I feel like it is very similar to my product and makes me want to give up entirely, but I will not give up. I feel like there is a true value in having substrates and agentic first approaches that Obsidian doesn't have and can't achieve because they are opinionated in a different way, and also i can pivot to solve the problems that I am facing with agentic engineering, as for them their vision is already pre-determined. I am still trying to nail the vision of backlog-mcp, which badly needs to be renamed at this point lol. I was facing lots of issues with context engineering, it didn't turn out to be really useful and with organizing the tasks, i have almost 1000 tasks and artifacts in my personal backlog and the number is only growing, without proper organization this tool loses the value and i might end up ditching it entirely, which is something that i don't want, but this is a clear signal that i need to improve the ergonomics. Somehow automatically organize items much better. For example even thinking about this right now one idea i have is so that when listing epics or folders it should come with some kinda default routing strategies, like this kinda tasks should arrive here and so on... so that folders natively come out of the box with some kinda pre-determined routing/organizing strategies out of the box. Pretty much decided ahead of time instead of after the fact, because after the fact organizing is quite complicated and slow, i have to think where each task needs to go, where each epic needs to go and so on. But I have to think about this a bit more, but this is me writing this article and actually an idea hitting me about the problem that has me burning right now. Why am i not using backlog mcp as much as i used to when i started at first, i was using much more heavily, because it lacks certain ergonomics that breaks down once you cross certain amount of artifacts inside the backlog. And this is only the beginning, i want to somehow integrate with Agents without coupling, i might create an substrate for agents, but i don't know how. I don't want to become another agent orchestrator, thats not my goal, i want backlog-mcp to be agnostic, but i want correlation of tasks and agents to be deterministically persisted somewhere... and still don't know how to solve this yet. Maybe this is another reason why my adoption is slowing down. Ideally i want to watch what agents are doing and working on via the backlog mcp viewer. Lots of difficult and nice to have problems, but i am kinda stuck right now and as a result, i am working on some other things, thats what i do when i get stuck... i try to clear my mind and neglect the projects that i love, and move on to something else thats easier...

this is an engineering and personal vision, pain article, there is too much prose at some places, we could have some interesting snippets here and there, or bold statements and bullets and explanations, and stuff like that... we need to keep prose where matters but we need some bold eye catching statements here and there and explanations as prose, it should not read like an essay.

in some areas i want like engineering options, like almost like proposals Option A vs Option B, explaining the options and then explaining the rationale why i chose some architectural or philosophical choice based on the timeline and work that i did, you can see and mine that information. Like why would an engineer read this article? besides the philosophical values and tenets and my vision, i want engineers to take away actual hard technical engineering decisions that i made in the code, you can see architecture, substrates, what it means to be building an agentic first tool. I didn't want agent to know about all the core types thats where my idea of substrates arrived, i didnt want 15 tools so i had to create a core creation and update tells you about the exact shape of the substrate so that agent context doesn't get bloated with all the types, the agent sees the types and full substrate shapes only when it needs to. This is one of the core integral ideas when building agentic products, you need to make sure that context is explored on-demand and lazily instead of eagerly pushing too much context inside agent.

update the date to current date today for the article and make this latest article a featured article