Why I Built Bwocks

I didn’t set out to build a tool for companies. Though I work at one. And Bwocks can be useful at work.

  • I built a tool to help expand and connect a technology tree for a board game I was making
  • I built a tool to save time converting a list of pieces of website text into JSON. And to pull JSON values back into a spreadsheet.
  • I built a tool to organize and keep track of context I was feeding to LLMs again and again
  • I built a tool to quickly pull ingredients from a list of recipe articles
  • Or to generate 30 pieces of copy I could pull into another tool in bulk

In my experience, a lot of knowledge work, organizing in life, and hobbies looks about like this. There are some shared characteristics of this sort of activity. And some reasons why they don’t fit well into other tools or methods you could use to get them done

Some of the shared characteristics are that:

  • They could be done manually. But they’re a bit tedious and could move from taking a few minutes to potentially a few hours. Particularly if they need to be done with any regularity.
  • They aren’t huge projects. Data points may number in the 10’s to 100’s, not 1000+. So you don’t want to set up a process that’s too large or complex to deal with them. If you can code you can solve some of these with simple scripts, maybe a Python notebook. But setting up or remembering the environment, or essentially building a little tool may take longer than just jumping in with elbow grease.
  • Because they aren’t huge projects, you don’t spend long enough on them to remember little helpful details. The context between projects shifts every few minutes. The prompt to an LLM you used last week for the same task is nowhere to be found.

Some of the reasons other tools or methods aren’t quite right:

  • You could and probably do some of them manually. This sucks up your day on little tasks, shuffling and reformatting data, if you can’t find a better way.
  • There are AI spreadsheet-like tools like Airops or Clay. But they’re built for and priced for business. They lean into marketing and sales workflows more than general knowledge work. But most thinking and tasks like this aren’t built around leads. And they don’t need a dashboard. They’re great tools for what they’re built for. But their business focus shows in their functionality and pricing.
  • If you can code, some small scripts can do some of this. Perhaps a Python notebook to solve related things. But some of these tasks are too small. Or perhaps you only do them once a week. Or you have a number of similar tasks that have slightly different data or context. It’s the environment, context, and keeping up with all the little parts that makes this route inefficient.
  • There are Google Sheets plugins that allow you to pipe cells to LLMs. But that’s about all. And after trying several I’ve yet to find one that isn’t buggy, doesn’t crash, doesn’t show ads, etc.
  • Finally, LLMs can do parts of this. But in a chat format you both lose the context that worked last week pretty quickly, and the output still needs a bit of tweaking in another program. For straight up research or ideation on a single output chat is the way to go. But I’m talking about series of tasks that crop up so often.

I’ll dive into more of my thinking below. But basically I built something because the tools that were built for companies were getting in my way.

I think with LLMs. I organize with spreadsheets.

I use LLMs the way some people use a notebook:

  • To think things through
  • To try ideas
  • To rewrite something until it feels right
  • To get ideas on how I could rephrase something
  • To write something that’s mostly right so I can take it the last mile
  • To explore half-formed thoughts

And I use spreadsheets because they’re the quickest and lowest barrier to entry way to:

  • Organize ideas: sort, filter, structure
  • Compare apples to apples
  • Spot patterns
  • Build little systems (ideally, and this is where Bwocks really comes in, there’s an observation or a transformation of some data applied in bulk)

But somehow, these two things never really worked together.

Everything felt awkward and fragile

When I tried using LLMs with spreadsheets, the options were… not great.

  • Google Sheets extensions were buggy and slow
  • Prompts broke randomly
  • Context lived in comments, notes, or my clipboard
  • Every experiment meant more copying and pasting
  • LLMs themselves can take in spreadsheets. But it’s disjointed. You end up with the “last mile” problem with format (more on this later)

Most of the time, I’d end up bouncing between:

  • A chat window
  • A spreadsheet
  • A notes app
  • A folder of files

It worked — but it’s clumsy. It’s annoying. And it’s a waste of time. It feels like 2010 (and not in a good way).

Context was always slipping

The real frustration wasn’t “passing information to an LLM.” This is the new “googling.” It’s muscle memory.

It was keeping track of context.

Every time I wanted the model to “remember” something, I had to:

  • paste it again
  • reword it slightly
  • hope I didn’t forget an important detail

Every time the context of my task at hand shifted, the same (of course we don’t use LLMs for only one thing).

Personas, rules, PDFs, notes, spreadsheets — they were always somewhere, never in one place.

And once context starts drifting, the little details in outputs start drifting too.

I wanted a way to say:

“This is the context. It has a name. I can reuse it. I can swap it when I need to.”

That’s where Context Bwocks and Towers came from.

The annoying last step

Even when LLMs give you something useful, it’s usually not done.

It’s:

  • almost the right format
  • slightly awkward
  • not quite structured
  • close, but not ready

So you fix it. Manually. Again.

That last step is small, but it adds up — and it breaks momentum.

AI Cleanup exists for that moment.

Right-click, fix it, move on.

Also: the internet isn’t always there

This part is very practical.

I’ve been on a lot of flights.

I’ve stayed in places where Wi-Fi technically exists… but not really.

In those moments:

  • cloud tools don’t work
  • chat tools stall
  • local models are the only thing that run reliably

Most AI tools just assume you’re always online. And there are times when you have THE MOST time to think through things and you can’t effectively without access to your tools.

I wanted something that still worked when you’re not connected.

So I built the spreadsheet I wanted

Bwocks is the tool I kept wishing existed.

A place where:

  • LLMs feel like part of the spreadsheet
  • context is explicit and reusable
  • cleanup is easy
  • things work locally
  • experimentation feels lightweight again

I use it every day.

If you think with LLMs, like organizing ideas in grids, and hate fighting the “last mile” of AI output — this is probably for you.

That’s really it.