LISTENING POST / FIELD GUIDE FOR NEW OPERATORS
You don't need to know how it works.
You need to know how to ask.
Four short channels teach the actual mechanics behind AI coding — prompting, tools, agent loops, context — with live demos you can poke at. Below that, a running log of what's genuinely free to use right now, rechecked on a schedule so it doesn't go stale.
live signal — move your cursor over it
CH.1 PROMPTING
The model can't read your mind. It reads your words.
Most bad results trace back to a vague ask, not a weak model. Type a real prompt on the right and watch what gets flagged.
- Name the concrete outcome, not the vibe
- Give constraints: format, length, what to avoid
- Show an example if the shape matters
- Start typing to get feedback.
CH.2 TOOLS & FUNCTION CALLING
This is how a model actually does things, not just talk.
You give the model a schema describing a function. It doesn't run the function — it decides when to call it, and hands back structured arguments for your code to execute.
- The schema is a contract, not a suggestion
- The model only ever returns arguments — your code runs the real thing
- Results get fed back in before the model answers
{
"name": "get_weather",
"parameters": { "city": "string" }
}
CH.3 AGENTS & LOOPS
An agent is just this loop, run until the job's done.
Think, act, observe, repeat. No new magic over Channel 2 — just the same tool-calling step wired into a loop that keeps going until the task is finished or it gets stuck.
- Each cycle re-reads the result before deciding the next move
- Loops stop on success, a limit, or a human stepping in
- This page's "what's new" log below is refreshed by exactly this kind of loop
CH.4 CONTEXT & MEMORY
The context window is a desk, not a filing cabinet.
Everything the model can "see" this turn — instructions, chat so far, open files — has to fit on the desk at once. Add files below and watch the desk fill up.
- Once it's full, the oldest material gets summarized or dropped
- That's why long sessions get compacted, not remembered word-for-word
- Durable facts need to live outside the window — a memory file, not the chat
System prompt reserved. Add files or conversation to fill the window.
FREQUENCY LOG
What's actually free to use right now.
Curated, not exhaustive — a running log of free-tier and open AI coding capability, rechecked on a schedule by a research agent so this doesn't calcify into a 2026 relic.
SIMULATED CODER
Talk to a simulated coding assistant.
Scripted, not a live model — there's no API behind this, nothing you type leaves the page. But it's built to answer the way a careful one would: plain language for questions, and a visible think → act → observe trace when you ask it to actually fix or build something.
tip: press / anywhere on the page to jump here
OPERATOR LOG
Milestones, tracked locally.
Saved to this browser only. Nothing leaves your machine.