June 11, 2026
AI plans, commenters panic
Build a Basic AI Agent from Scratch: Long Task Planning
Coder says he built a smarter AI helper, commenters say the real bug is Medium
TLDR: The tutorial shows how to make an AI assistant handle longer jobs by giving it a notes page and a task list. Commenters, however, were far more fired up about AI hype, sloppy buzzwords, and the audacity of posting a coding guide on Medium.
A developer dropped a tutorial on Medium showing how to make an AI helper stick with bigger jobs instead of giving up after one tiny step. The idea is simple enough for non-experts: give the bot a scratchpad to jot down its thoughts, plus a to-do list so it can break a big job into smaller ones, track what’s done, and rethink the plan when things go sideways. In plain English, it’s about turning a chatty AI into something that can keep working without needing constant hand-holding.
But the actual fireworks were in the comments, where readers seemed less interested in the code than in roasting the entire vibe. One person rolled their eyes at the endless wave of “AI agents,” sarcastically asking how this would help with, you know, climate change and child starvation. Another went straight for the language, saying the phrase “long term planning” is being abused so badly it would make academic researchers wince. And then came the surprise villain of the thread: Medium itself. Multiple commenters basically acted like the platform was the biggest crime here, groaning about code tutorials being posted on a site they say is terrible for code formatting.
So yes, the article is about making AI more organized. But the comment section turned it into a spicy little referendum on AI hype, buzzword inflation, and why people are still publishing code on Medium in the first place.
Key Points
- •The article is a continuation of a series on building a basic AI agent from scratch and focuses on long task planning.
- •It says the prior version of the agent could already find files, read and write files, run bash commands, and access web content.
- •The article argues that conversationally trained LLMs tend to stop early on long, complex tasks and need explicit planning support.
- •It identifies key planning abilities for the agent, including goal understanding, task decomposition, progress tracking, replanning, and completion checks.
- •It introduces two in-memory tools—a scratchpad and a to-do list—to support planning and execution, and notes the code is available in a GitHub repository.