My Wise Buddy and I #1: Confessions of a Link Hoarder

I'm an ordinary office worker in a decidedly non-tech industry. Not a developer — I don't write software for a living, or as a hobby, or at all. My relationship with code has always been like my relationship with the gym: deep respect, zero attendance.

But I've been quietly obsessed with AI since ChatGPT blew up in 2023. Not building anything — just following it, with the devotion of a sports fan who has never once touched the ball. Newsletters on the commute, social feeds over lunch, YouTube explainers at night. Three years of watching, from the stands, a field that moves faster than anything I've ever seen.

Then, this year, work gave me an unexpected reason to open a terminal: I got to try AI coding agents — Claude Code and Codex — on real tasks. That's when I discovered these tools have changed what someone like me can actually build. This was a dangerous discovery, and this series is about what I'm doing with it.

It begins, as all honest projects should, with an embarrassing confession.

The write-only archive

For years, my entire "knowledge system" has been a single Telegram chat. Just me, forwarding links to myself. A private museum of everything I ever meant to learn.

Every useful thing I found — a YouTube video on agent architectures, a prompt technique, a benchmark screenshot, an article on some new model — went into that one chat. Thousands of messages by now.

Number of times I went back and used any of it: approximately zero. The museum has no visitors. Including the owner.

You probably have a version of this. A bookmarks folder you never open. A watch-later list measured in days of runtime. A notes app with one giant note called "misc." Saving feels productive — every save prints a tiny receipt that says "I'm learning" — but nothing compounds. Information goes in; nothing comes back out.

I call it a write-only archive. And the failure isn't the collecting — collecting is easy, fun, and frankly I'm excellent at it. The failure is that collecting was the only step that ever ran.

What changed

The honest answer: capability, not motivation. Motivation was never the bottleneck — mine has been running at 100% for three years. It just spent all that energy collecting more links.

I always knew the fix in theory — organize the saves, distill them into notes, connect the notes, review them. Every productivity guru says the same thing. But as a manual process, on top of a full-time job, it's a diet plan: correct in every detail, and abandoned within two weeks.

Working with Claude Code and Codex showed me a different shape for the problem. These agents can read files, write files, run scripts, and follow written rules — reliably enough to matter. Which means the boring parts of a knowledge system (filing, formatting, summarizing, cross-linking) don't have to be my habits anymore. They can be someone's job. The someone doesn't have to be human — which is lucky, because the only human available has a well-documented track record.

So the project is not "become more disciplined." Three years of evidence says how that one ends. The project is: build a system where AI agents run the pipeline, and I only do the parts that need my judgment.

Meet WIBBY

I named it WIBBY — short for Wise Buddy. Yes, I gave it a name. If you're going to talk to a thing every day, it deserves one — and it changes the relationship: WIBBY isn't software I use, it's a colleague I delegate to. A colleague who works nights, never sighs, and has seen the state of my Telegram chat and stayed anyway.

WIBBY is not an app I'm going to sell. It's a personal knowledge operation with three components:

• Plain markdown files as the single source of truth — an Obsidian vault, organized so both a human and an AI can navigate it
• AI agents that operate the pipeline — capture, file, distill, connect, resurface
• Me as the audit layer — the agents draft, I spot-check samples and decide what ships

The pipeline WIBBY needs to run, sketched at the altitude I understand it today:

Stage	What happens	Who does it
Capture	Saves land in one inbox (yes, Telegram stays — it's where my habit already lives)	Bot
Organize	Each item gets filed into the vault with metadata	Agent
Distill	Long raw saves become short atomic notes in my own words	Agent saves a draft; I spot-check samples afterward
Connect	New notes get linked to related existing notes	Agent proposes
Surface	The system brings things back — briefings, resurfaced ideas	Agent

That last row is the whole point. A knowledge system that never brings anything back is just a write-only archive with nicer furniture.

Three rules I'm starting with

1. Files over apps. Everything lives in plain markdown on my disk. AI agents are shockingly good at grepping and editing text files, and no vendor can take the vault away from me. Apps come and go; a folder of .md files has no opinion about anyone's business model.

2. AI operates, human approves. I don't want to do the filing, but I also don't want an AI quietly rewriting my notes while I sleep. The default split: agents handle mechanical steps autonomously, and anything that changes meaning lands as a draft that I audit — a random sample every week, not every note. Trust, but verify — especially when the thing you're trusting types faster than you read.

3. It must survive neglect. I have a job. Some weeks I'll have zero spare hours, and the system has to keep running — or at least fail loudly — without me. This rule has already earned its place: after a reboot, a Windows Task Scheduler default silently kept part of this stack from coming back, and I only noticed by accident. I wrote that postmortem here. Nothing teaches you the value of "failing loudly" like a system that fails in perfect, dignified silence.

Where this series is going

The plan, loosely: get WIBBY working as a real assistant first — capture to surfacing, running daily on spare-time maintenance. Then push further. My long-shot ambition is to grow this from a personal assistant into something closer to a one-person AI operation — but that's a later chapter, and I'd rather show it than promise it. The internet already has enough grand plans from people whose main asset is a Telegram chat full of links.

If you've ever hoarded knowledge you never used, follow along. This series is me finding out how far an ordinary office worker can get — with a wise buddy doing the heavy lifting.