Experimenting with Rabbit Remix and AI for strategic idea generation

I've been experimenting with chaining AI and other API tooling together to provide valuable decision-support mechanisms for strategy. Here's what I have so far.

Photo by Larisa Birta on Unsplash

Many AI capabilities are now available through APIs, which can be accessed through easy-to-use user interfaces - even to integrate with your existing applications.

Many tools are available to help you daisy-chain AI API calls with other services to create more sophisticated combinations.

As I highlighted in the previous post, the downside remains that this low technical effort barrier means your invention is likely easy to replicate.

The most famous examples are Zapier and IFTTT, but many more are on the market. I have chosen a more experimental option as it has a few more exciting features the creator is experimenting with, and it is capable enough to warrant more attention:

Introducing Rabbit Remix

Check out Rabbit Remix here - I recommend watching some of the posts and videos there.

Rabbit Remix

You could say Rabbit Remix is a technical demonstration of sorts, one of a series of very functional technical demonstrations that Ryan Robitaille has built on the pathway to something even more remarkable - read about his current focus: Rvbbit

Ryan's videos, blog posts, and Twitter posts are intriguing, and if you get hands-on, you will find them essential to understanding how to work with Rabbit Remix.

Just how powerful it is, even with Ryan’s impressive examples, is hard to grok without trying it yourself, so if what you see piques your interest, let's get into it!

Getting started is pretty straightforward - here’s a set of steps I followed to get up and running:

1. Download the Rabbit Remix server:
   https://www.rabbitremix.com/downloads/

2. Run the server:
   java -jar <name of the file you downloaded>
   (It assumes you have Java installed and in your PATH. If not, a Google search can lead you to simple instructions on how to do this)

3. Create and enter a Github API key so you can save your flows… Rabbit Remix can store what you create as Github Gists

   1. https://github.com/settings/tokens/
      - Generate a new Token (Classic)

   2. Set an expiry (suggest not unlimited)

   3. Set a scope - Gists are adequate; don’t add scope beyond what is needed.

   4. Press ‘Generate Token’

4. Enter your Open AI key

   1. If you need to generate or access your API keys, you will find them here:
      https://platform.openai.com/ (OpenAI > Products > API login)

   2. To generate a new key: Press ‘+ Create new secret key’ and follow the instructions.

   3. Copy the secret key and store it safely in a password manager (it won’t be shown again)

5. Go to http://localhost:8888/ in your browser. Rabbit Remix loads and presents a window.

   1. Select ‘Saving & API keys’ to switch to a tab with some configuration fields.

   2. Enter the keys from Steps 3 and 4 into the corresponding fields.
      Note: you will need to return to this window explicitly to save - press G or ? to have it appear again.

I recommend reading this particular post from Ryan about one of the most fundamental concepts underpinning Rabbit Remix, and that is how to use the text remixing features of Rabbit Remix:

https://www.rabbitremix.com/pre-release-overview-part-2/

Picking an interesting experiment: Strategy idea generation

I work in consulting these days, focusing on strategy and working with C-level executives. Research and analysis often require applying structured thinking and generating possibilities. I wanted to test the ability of AI and a workflow system to speed up the process dramatically.

Of course, I am worried about the AI hallucination or other side effects. My hypothesis: If we can break the process into steps, there will be transparency on how it works, and the blast radius of any individual step will be less and can be easily sanity-checked.

I’ve been using quite a few AI-enhanced tools in the last 12 months, and those that seem to cross the threshold of reliable tools I reached more than once appear to use the orchestration strategy to get better results.

They are likely also powered by LLM (“Large Language Models”) instances, for which these companies have used their curated datasets and extensive fine-tuning rather than generalised consumer models like ChatGPT, Claude, Gemini, etc.

For this experiment, OpenAI’s ChatGPT is more than adequate.

Generative AI as the small gear that turns a larger one

For now, I’ve selected a relatively straightforward strategy heuristic to test with. It’s more of a market and product positioning tool than a sophisticated approach to strategy development. Still, it’s interesting in its own right; it is sufficiently complicated to be a helpful test and sufficiently simple for the reader to follow what is happening.

A description of the steps involved in researching a list of competitor companies and generating potential strategic positioning is here - I recommend reading through and imagining a workflow that uses generative AI prompts to do each step and to feed some of the results, such as lists of attributes of competitor companies into subsequent steps:

Alex M H Smith: How do I get my competitors to write my strategy?

The method, as you can see, might generate many plausible but ultimately unviable ideas. But amongst those, maybe something useful or, more likely, the germ of an idea that may inspire people to identify viable ideas.

I wouldn’t have the machine follow these steps, spit something out and use it as a strategy. The point of this post is not about ceding strategy over to a machine. Still, I could imagine it might be an interesting input to discussions on a strategy day or similar efforts to explore new opportunities.

Interestingly, this approach pulls significant amounts of data together in a very short time—the sort of research input that wouldn’t be possible without significantly more time and human power to prepare. The effort would not have been low enough to do this work for most companies, and now it can be ready to use within minutes and, more importantly, continue to be tweaked and interacted with by people as part of exploring ideas.

To me, this seems like a particularly good candidate for the use of current-generation generative AIs—we are less strict on complete accuracy, and the purpose is to help illuminate opportunities and potentially inspire new thinking. This is an example of AI as the small gear turning to a larger gear, the critical reasoning and novel invention capabilities of people.

In a future post, I may share some of my experiments with Rabbit Remix, but honestly, it's fun to try it out for yourself. You can use the example strategy process as your goal or choose your own AI automation challenge.

If you sign up for an ElevenLabs account, you can talk to Rabbit Remix to provide your AI prompts and have it describe what it’s doing. There are a few other cool things you will notice if you examine the default flow Ryan has loaded:

1. there is the ability to define subflows for some modularity and build up sophisticated concepts.

2. he has used the System pragmas in the AI prompts to define some intelligent agent-like behaviour and explained to these agents how to edit the canvas. You will see demos of Ryan asking the agents to do some tasks, and they build up their own flows on the canvas. This is stuff only really seen in movies and you can use it to do practical work.

Happy remixing!

For comparison, I used AWS’s PartyRock AI app to generate a similar concept tool to replicate Alex M H Smith’s strategy heuristic. That example is here:

https://partyrock.aws/u/wioota/_FeCbLG6r/Business-Differentiation-Strategist

Have you experimented with generative AI in combination with workflow orchestration tools? What experiments did you do? What were the results? Share your experiences in the comments.

---

As a humourous aside, Alex M H Smith’s approach reminds me of this video:

https://timharford.com/2024/02/the-secret-to-finding-the-best-idea-ever-first-think-about-the-absolute-worst/

---

Check out my monthly livestream for CTOs with my co-host Noah Cantor here:
CTO Life Line Youtube Channel

If you would like my help as a coach or consultant: