I’ve spent almost a decade working with design systems, often obsessively. Much of that time has gone into tokenization, fine-tuning and building controls for almost every variation I could anticipate. I wanted to be able to express a design precisely, down to the smallest detail, using the least amount of tuning parameters.
Lately, I’ve started letting go of some of that control. I’m putting more trust in a clearly expressed design intent, the principles I carry across projects and an agent’s ability to work with both.
That change has made me reconsider where design happens. Content, brand direction and design decisions are closely connected. As one changes, the others often need to be reconsidered. I increasingly want to work wherever I can bring that context together.
Keeping the context close
In my own workflow, the tools have fairly clear roles. I use Notion for project management and task orchestration. Google Drive holds files, images, contracts and the administrative material around a project. My Claude Design artifact holds the design system and the local setup where I explore ideas.
These are parts of the same process. An updated brand identity in Drive may change the direction of a page. Trying a layout may reveal that the content needs a different structure.
The design helps me understand the brief, and the brief gives me a basis for judging the design.
Keeping that material organised and accessible makes it easier to carry decisions between environments. I still need to know which version is current and which decisions I’ve accepted. But I can ask the agent to work from that shared context instead of explaining the project again each time I move.
Room to experiment
A local workspace gives me room to try things: a different layout, a custom dashboard, a small interaction or an idea that may never reach the website. I can keep alternatives, compare them and choose what deserves more work.
I feel comfortable experimenting because I know there is a practical route from that work into Delta. The theme works with HTML, CSS and JavaScript, so I can explore a design using the same materials I’ll use on the site.
That continuity matters. When an experiment works, I can bring selected parts into WordPress without reconstructing the layout through a separate library of builder elements.
There is still work to do before it belongs on the site. The experiment needs to fit the templates, shared styles and real content. I need to review it at different screen sizes and check how its interactions behave.
Having a route to implementation gives me freedom to explore; deciding what is ready remains part of the design work.
Letting intent guide the work
For me, the harder adjustment has been letting the agent make decisions I used to specify myself.
I spent years anticipating variations and providing tokens and controls for them. That precision was useful, but maintaining it also took up a considerable part of my attention. I’m now putting more of that effort into crystallising the intent: what a page should communicate, how its parts should relate and what should remain consistent as it changes.
Some principles stay with me across projects: how I approach hierarchy, spacing, typography and visual emphasis. I’m trying to express those principles clearly enough that an agent can use them in situations I haven’t specified in advance. Tokens and shared rules still help. I’m simply becoming more comfortable seeing how the agent applies them before deciding what needs adjusting.
If I want a three-column grid to become two columns, I can explain the change and why I want it. I can then judge the spacing, balance and behaviour on smaller screens. I don’t need to settle every value before I have something to look at.
That is the trust I’m developing: enough room for the agent to interpret an intention, with my judgement still involved in the result.
What I still need on the website
Just the right amount of higher-level control. I also want a place to draft ideas in the context of the website itself. Delta’s front panel and Sketchbook give me that option.
The Sketchbook opens inside the front panel. Its boards use the site’s own styles, so I can explore a different hero or palette without changing published pages. A board can become a draft page; applying its experimental styles to the website is a separate choice. The work can move back into the agent workspace, too. A connected agent can retrieve the board’s HTML and CSS, allowing me to continue exploring it in my local Claude session.
Sometimes that is where I want to work. At other times, I want to stay close to the website, sketch an idea or change a few words directly in the front panel. Someone else may prefer Paper, Figma or a familiar visual editor.
I value having that choice. The place I work can change with the task, while the project’s context and design intent give the work continuity. I’m still learning how much to specify and how much to leave open. Being able to try, inspect and revise makes that a useful question to keep exploring.
