Infact I often find the highest quality thinking and work for me happens when I am away from my desk and placed in a more mobile environment. I have been working in Robotics for the past few years with a codebase where robot behavior was buried deep in 1000s of lines of imperative code. One of the biggest contributions to the space was developing a behavior-tree/compositional-recipes style framework for writing robotic behaviors. I still remember driving to work noodling the thought about - how Javascript's promise chaining looked so elegant and what a similarly flavored robotic DSL might look like. After 2-3 design attempts, I landed on an elegant architecture in writing robotic behavior. New robotic skills could be developed in 40 lines of YAML code, instead of 1500 lines of intricate python. As a side effect, the concise Skills, Recipes YAML interface is also extremely parsable by an off-the-shelf LLM - we got VLA-style robotic motion without any model training. Incredible to see.
If I think backwards about my process - I dare say that all of the core breakthroughs ideas - late-binding of resource, declarative langague, javascript style promises - were all separate ideas that all came during walks, discussions (with other engineers, with LLMs), during drives, some in the morning shower. Almost every idea that wrote down sitting at my desk - I either re-wrote, re-factored or deleted outright. There is an order of magnitude difference in the quality & value of work I produce, when it is accompanied with deep thinking and brainstorming. Most of the work I produced directly sitting at my desk - would qualify as things that did not require/get deep thought. Don't get me wrong - most of my day is spent at the desk. Sitting at my desk is great for executing on something that I already know, however to develop new directions I always need to mull the problem and go do something else. Then at some point the computations in the brain seem to timeout and yield an interesting question or thought.
My broad thesis is this: Executing on an idea has become incredibly cheap. The differentiation or onus has shifted - to the thinking, brainstorming and exploration process. For me that often happens - on a run, or on a long drive, morning commute. Whenever a good question/thought arises I would like to be able to immediately register, jot it down, brainstorm with a fellow engineer. This gap is where thinking fades. Ariadne is my attempt to close that gap, for myself.
Ariadne - runs on your local workstation, on your code repo. Dial it from anytime, anywhere & brainstorm with an informed Architect on the code repo. Launch structured investigation tasks using your local friendly coding-agent (Claude, Claude, ..). Refine your ideas, reframe the ideas and change the framing in a conversation. If you arrive at a concrete idea generate an implementation brief for your coding agent with all the findings, context, problem framing. When you reach your desk, there is an implementation brief ready to go.
Ariadne is built on Daily's Pipecat framework for voice-agents. It works incredibly well.
What I'm uncertain about: Is this just a quirk of my mode of operation? Is this useful to anyone besides me?
Looking for: 10 developers who might be interested in trying this out and sharing thoughts or feedback - by using it in their actual workflow. For the first few developers, I will set you with my API_KEYs.
Ariadne Demo: Implementing a feature: https://www.youtube.com/watch?v=MKwRE-1cQdI
Ariade GitHub: https://github.com/RavindhranSankar/ariadne/
Scheduling: https://calendar.app.google/xC41ED7tz9mkozdu5
0 comments