40 vs 7 Hours for Analysis & Strategy (Without or With AI)
At Tigre Rossa, we’ve been using AI in our day-to-day work for quite some time now, but mostly as a collection of separate tools for separate tasks. Sometimes it helps us write a text, speeds up research, generates an image, video, or checks a piece of code.
The quality of generated visual materials provided by modern tools is advancing with an incredible speed and the outcome becomes always more in line with our design taste and quality metrics. One of the few remaining issues was the time wasted for endless onboardings: each time we were providing AI with the whole design system context. And then passing to actual task, and new prompt.
Now we decided to try a different approach and gradually train our own digital assistant. That became the focus of
Skills isn’t just another chatbot, it acts like a trained, proactive colleague with a rich, well-structured database always at hand. Our Data Artist described it as: “Onboarding a new colleague, but only once”.
The workspace looks like a dream for anyone who loves perfect order. It has a structured collection of files, which are divided into two main sections:
- a brand description;
- visual principles;
- the design system;
- working instructions;
- example prompts;
- links to supporting materials;
- brand assets;
- references;
- reusable templates.
Our Data Artist created a dedicated folder containing everything that helps Claude understand Tigre Rossa’s visual language: images, PDFs, Figma examples, brand elements and a wide range of visual references.
We started with our own social media content. For our graphic designer creating a single post doesn’t take that long, but if you look at the entire workflow, you realise how many small decisions are involved across the process:
- Choosing the topic.
- Structuring the story.
- Selecting the format.
- Writing the text.
- Building the composition.
- Creating the visuals.
- Checking brand consistency.
- Going through several iterations.
Perhaps the biggest mistake is expecting perfect results from the very first attempt. Every generation is an opportunity to spot patterns, repeated mistakes and fix them directly in the Skills instructions. With each iteration, the repeated mistakes become less frequent.
Gradually, the Skills stops being just a collection of instructions and starts evolving into something much more interesting: a living knowledge base for the entire company, which soon might be capable to act almost on its own.
We’re still a long way from saying, “AI can do EVERYTHING on its own.” But even at this stage, we’ve already noticed a couple of improvements.
We’ve eliminated the need to repeat the same context over and over again, and consequently, saved time on this infinite loop.
Our designer spends less energy on repetitive preparation and can move faster towards the parts of the work where experience, judgement and her creativeness are required.
Perhaps most important, we’re gradually building a shared knowledge base that grows smarter overtime and alongside the team.
Now we write a prompt, and we have a ready-to-use layout for our posts: fonts, composition, colors, contrasts are right. Designer than needs to add a finial creative human touch. Choose a more authentic illustration to fill into the layout, for example. Generated directly within the Figma, where both Claude Skills (for layouts) and Chat GPT (for image generation) are connected. Or she decides that this time composition should be bolder, or less perfect, with some accidental drawings on top.
The Skills is evolving together with the agency and in a few months, we’ll be especially curious to measure what has changed.
Has routine work decreased? How much faster is onboarding a new client? Has content production become noticeably quicker, say, we create 10 more posts per month with less effort? And perhaps the biggest question of all: Is it possible to grow an AI colleague who understands your company almost as well as someone on the team?
It looks that way. But more answers in the next chapter, which probably will nourish new, even more interesting questions.