Where we're going, we don't need vibes.

I’m not a fan of the term “vibe coding”. It doesn’t do enough to describe the process of working with agentic tools to build software, and I suspect it’s also contributing to the diminishing value placed on user experience and product designers and builders within some corners of our industry. I was much more into the idea of Home-Cooked Software and Barefoot Developers, but alas those terms were far less catchy!

Science, not art

Design has always been done better when paired with methodology and process: Lean UX, design sprints, the double diamond, typography scales, grid systems, Hick’s Law, yadda yadda yadda… are all examples of structured approaches to design. In the same way within teams, good products are built following structures and processes like Agile, Kanban, Scrum. Without these structures, ways of working can quickly become unsustainable, jobs to be done unmanageable, and the product quality and user experience degrades. Things get sloppy, fast.

As technology evolves, perhaps new methodology and process to support us to design and build may also emerge.

Messy… by design?

In the current state of play, I can run any number of conversational threads with agents across different tools at once (eg Claude + Cursor + …), spin up a side chat or five within them, work with multiple AI agents and models in parallel across separate CLI terminal windows and as well as desktop apps and for bonus points, keep chatting to Claude Code via remote control on my mobile phone while I’m on the loo.

I’m feeling a growing unease, where the tools themselves seem to almost encourage a messy, exploratory, stream of consciousness, “vibey” approach, and in the process entice you to drift away from an intentional decision making, structured thinking and methodological workflows, and instead slip into easily-distracted, chase-the-shiny-thing type behaviour. So far I’m successfully holding the line at not coding while pooping, though I observe myself slipping in other ways often! The tooling is feeling increasingly messier, and my assumption is people’s workflows are starting to mirroring this.

Sure we can make rapid, early progress, with the heavily compressed time it now takes to go from 0 - 1 delivering a dangerously strong dopamine rush, but where are the guardrails to help ensure we are designing and building coherent experiences, in coherent, sustainable ways? Seemingly by design, the tooling is playing into the whimsical, chaotic, stream of consciousness nature of humans. Feature? Or bug?

Seeking intentional approaches in the ever-growing sprawl

I’ve challenged myself to seek systematic and methodical approaches to help me design and build intentionally within agentic tooling.

I’m comfortable knowing that in six months time this post will either read like gibberish, or like it was written by a time traveller from the distant past, or that for some reading this it might already seem like either really primitive or advanced thinking. This post is just a snapshot of where my head’s at, today.

My friend Josh Crowley has a concept of “recipes” which I love, but I’ve been thinking about formulas (he says potato, I say tomato). Formulas are the emerging sets of repeatable patterns I’m trying to codify into my process when working with agents. They can be used in isolation, or chained together for deeper and more complex problem spaces and output.

They often begin as tasks I notice I’m consistently repeating manually, a common by-product of the cyclic, loop-based nature of working with agents.

The words I’ve used to name the formulas aren’t new, but wrapping a system around them is, at least for me.

Every formula requires two attributes to be potentially successful, and without them you’re farming out too much of the decision making to a non-deterministic technology, and where’s the fun in that?