"Choose Your Own Software": How to write new tools and get work done
Dear Operators, try doing something many times a day across GTM systems. How do you build solutions faster? Enable your users. Read: "Everything Starts Out Looking Like a Toy" #236

Hi, I’m Greg 👋! I write weekly product essays, including system “handshakes”, the expectations for workflow, and the jobs to be done for data. What is Data Operations? was the first post in the series.
This week’s toy: the Pebble Watch is returning - this time to open source. It’s a great opportunity to make simpler, single-purpose wearables that don’t do everything (like a watch for kids). Edition 236 of this newsletter is here - it’s Feb 3, 2025.
Thanks for reading! Let me know if there’s a topic you’d like me to cover.
The Big Idea
A short long-form essay about data things
⚙️ "Choose Your Own Software": How to write new tools and get work done
If you walked to Waldenbooks (a long ago bookstore chain) in late 1984, you might have been visiting there to pick up the last entry in the popular “Choose Your Own Adventure” series: “Supercomputer”, by Edward Packard.
The CYOA series – if you’re not familiar with the concept – was a series of book as hypertext reads where depending on your choices, you reach a different branch or ending. The idea of a “choose your own adventure” form of entertainment was not new. Zork, a turn-based computer exploration text game, had been released in 1980.
But Choose Your Own Adventure was different because it took this innovation to a new place: stuffy old books.
Here’s the back cover of “Supercomputer”:
You’ve won a computer programming contest. Now you’re the lucky new owner of a Genecomp AI 32 computer named Conrad. You read the operating instructions carefully and turn on the power. Suddenly Conrad begins to talk. He already knows everything about you! your new supercomputer is a genius. You can do almost anything with him!
If you ask Conrad to help you make a million dollars, turn to page 4. If you ask him to help prevent war, turn to page 14. But be careful. There are dangerous people who will stop at nothing to get their hands on Conrad. You might end up lost in outer space, escaling a revolution in a hot hair balloon — or receiving a brain implant that could turn you into another Einstein!
Why is this important? We’re on the verge of a similar breakthrough in software, where creation tools make their way to mass adoption by regular people who are not computer programmers.
“Choose Your Own Software”
What do people want from software? A process, agent, or “magic” that helps them complete a task with a known output. Whether this is a commercial product or a series of reminders depends upon the Job to Be Done. You end up deciding “it works” or “I need a new tool” based on your approval of the output and the effort required.
When you’ve done something before, you have a template or a blueprint for doing it again. A macro, a sample piece of code, a document template: these are shorthand for “finished work” or “reusable idea”.
Generalize this idea as:
Repeat known process
Determine what input tokens are needed
Measure the result against a standard
And you have a thought process that seems a lot like the way LLMs repeatedly process information.
“Supercomputer” introduces the creative process of building software this way. First you introduce yourself, and then you ask it to complete a task. As the computer gets familiar with your preferences, it will better match the results you’re seeking.
This is a prompt from 1984, and it sounds eerily prescient to the workflow we’ve built into AI tools. Roughly speaking, the workflow looks like this diagram.
It’s simplistic, yes, and that’s the point. When we have conversations with software to build process, we build primitive interactions that can be built up into a framework or grammar for action.
A Tool for Creating Tools
Traditional software models start to break down, particularly for B2B, when the change that you want to build (or problem you want to solve) is too niche. Build for more people and you will optimize the effort to build the code. You’ll also get more enthusiastic users if you build features that people really love instead of the ones they tolerate.
In the Choose Your Own Software mode, however, you’ve got a tool machine that makes tools. Start having a conversation with your Conrad LLM and (with the right integrations) you’ll have a tool you can use pretty quickly.
Traditional tools are either completely open ended, letting you write your own code, or constrained to the existing UI (do it this way). For the sake of this argument, I’m calling API integrations “open-ended” too.
Software building is about to change for some segments of the market.
If a task is not already solved, your agent will ask you “what do you want to build today?” Through iterative prompts and existing building blocks, you’ll get to a solution much more quickly than an engineering team will build it for you.
"Based on what you've shared, here's what I think we're trying to achieve..." (h/t to Luke Tucker at Procore for the inspiration)
If the system responds like this, it’s a clue that the software process is iterative and might not look like what you imagined during your initial conversation.
This begs the question: how can you let people use tools like this and keep the internal software safe and someone close to its original intended use? One way is for engineering teams to build the primitive building blocks for their users and then string these together using familiar interfaces (Slack, Email, Chat, Voice). Once proposed apps are created, they need to be sanitized or sanity checked (probably by automated build tools).
Find the best way to do the next thing
LLMs are great at indexing data and guessing the “next best thing” that answers the question. We need to teach them to help users to find the tool that will solve their next step, either by finding an existing solution that does the job or by walking them through the steps to create a new building block.
The “job to be done” in this case is making it easy for teams to find the right next step and getting reliable output. It’s also making sure that they have the right tools to build software without compromising security and data for the organization … so these are a complicated balance to maintain.
Building is a Choose Your Own Adventure game after all, and we need to precompute some of the outcomes so that we can give the users some hints. (Bonus: we will get great new ideas from them that did not surface during “requirements gathering.”)
What’s the takeaway? We finally have the infrastructure and the tools to take the “Choose Your Own Adventure” idea into building software for the average office worker. Next? We need to determine the guardrails they’ll need to make this effective and not distracting.
Links for Reading and Sharing
These are links that caught my 👀
1/ There are always more channels - Rex Woodbury writes about AI and its effect on our common culture. One (maybe non-obvious) effect is that with more choices, it’s less likely you will be rooting for a team or watching a show that someone else is watching.
More choices are generally good (unless you are trying to pick a Thursday night show to watch with a few other people). This raises the need for algorithmic (or manual) curation of the best stuff. However, the economics of streaming services means they aggregate viewers who pay and rewatch the same old stuff (or forget they have a login). Perhaps the next model will look more like pay-per-series (TV rental or Sports PPV vs Movie Rental).
2/ Just Build Something - If you don’t see a product, build it! That’s the idea behind this hardware odyssey from Andrew Childs about creating a wearable for Type 1 Diabetes. This very readable summary suggests that LLMs and other tools lower the barrier to creating prototypes. (Good opportunity to build pickaxes for miners who want to build wearables.)
3/ Plain truth - Sometimes, making a very plain thing is genius. Plain Text Sports is a great example of this axiom. The site is a scoreboard and ticker for sports news and does it in the style of a 1980s BBS. It loads lightning quick and will work on any device, which is the point. Utility + speed FTW.
What to do next
Hit reply if you’ve got links to share, data stories, or want to say hello.
The next big thing always starts out being dismissed as a “toy.” - Chris Dixon