How far AI can get me within a week?

Outcome

• Full brand, website, and launch video shipped in one week.
• Clear per-phase scorecard — where AI is 5/5 and where it is 1/5.
• A reusable workflow: AI-first for research and code, hand-first for brand and video finishing.

AI scorecard

1. 1. Research — competition, vision, features

   Defining market needs, visual rules, main players, and the company's identity. Summing up the features and information architecture of the leading industry players. This is the area where AI is not great enough to tap into a highly specific industry built on human connections and semi-casual offline meetings.

   Using: Claude (competition skill + chat)

   AI rating

   Speed

   5/ 5

   Brainstorm companion

   3/ 5

   Skills

   3/ 5

   AI credibility / professionalism

   3/ 5

2. 2. Research — interview

   Finding out the company vision and goals, industry rules and leaders, and the team's best references. Interview is the part I would hardly hand over to AI, even though it is easy to do technically — researching the mood and vision of the team is a very human thing. With Claude I brainstormed interview questions on top of a questionnaire I have been using for years, modifying it for each case.

   Using: Claude (interview skill + chat)

   AI rating

   Speed

   3/ 5

   Brainstorm companion

   3/ 5

   Skills

   3/ 5

   AI credibility / professionalism

   3/ 5

3. 3. Brand

   Conversation-based clients like GPT are really good companions for brainstorming and idea development. Yet AI-based logo tools are not great in execution. I ended up creating handmade sketches, vectorizing and tweaking them in Adobe Illustrator.

   Using: Claude (chat), Adobe Firefly (logo generation), Figma Make (logo generation), paper + pencil + visual research + brain.

   AI rating

   Speed

   1/ 5

   Brainstorm companion

   5/ 5

   Skills

   1/ 5

   AI credibility / professionalism

   1/ 5

4. 4. Website — design

   The website design starts in Figma. The minimum initial inputs Claude Code or Cursor need to do a decent job are: 1) color scheme, 2) fonts (ideally three), 3) styles, 4) a variant-based UI kit, 5) information architecture, 6) texts and microcopy, 7) designs for all screens, 8) screen adaptation rules.

   Step 1 — colors and fonts chosen by hand. Step 2 — Figma UI kit + prototypes so stakeholders can approve the information architecture and info blocks. Step 3 — website design (a mix of Figma Make and my own hands). Keeping the file neat and tidy was on me too.

   Using: Claude (chat), Figma, Figma Make. Tested Cursor.

   AI rating

   Speed

   4/ 5

   Brainstorm companion

   3/ 5

   Skills

   3/ 5

   AI credibility / professionalism

   2/ 5

5. 5. Website — coding

   Feeding Claude Code this base through Figma MCP and coding section by section. The initial AI coding experience is scary-amazing. Though you need to make the first steps right — pass the main UI elements, instruct AI on how to read your files, and set up tokens.

   Using: Claude Code, Figma MCP.

   AI rating

   Speed

   5/ 5

   Brainstorm companion

   3/ 5

   Skills

   4/ 5

   AI credibility / professionalism

   3/ 5

6. 6. Website — review and changes

   If you have ever worked with AI, you know that making small changes is harder than creating something impressive from scratch. Figma Make or Claude Design can generate impressive results from zero, but if you want to move a button you can spend an hour trying to explain what you want. This stage is much easier if you know how to code.

   My personal combo for being able to fix things myself is Claude Code + Visual Studio + Git. At the moment of building this site that was it. Now I would add Lovable to the mix.

   Using: Claude Code, Figma MCP, Visual Studio, Git. Tested Cursor.

   AI rating

   Speed

   1/ 5

   Brainstorm companion

   3/ 5

   Skills

   3/ 5

   AI credibility / professionalism

   3/ 5

7. 7. Video

   I was sold on AI being easy for video. It really is not — in my experience it is crazy demanding on time, skills, and tokens. First attempts to generate a video from scratch gave unexpected results, and changing anything was even harder.

   I settled on this workflow: (1) generate a full storyboard; (2) generate image by image using clear references and descriptions written with ChatGPT — some images I had to edit in Photoshop before regenerating, because AI could not handle the task; (3) bring each image to life with a set of prompts — painful and slow; (4) do most of the color, speed, and final montage by hand in Adobe Premier. To keep the video less AI-like we also decided not to show faces.

   Would it be easier or faster to film it? Roughly the same, depending on the production team — with a solo video creator it comes out about equal.

   Using: Adobe Firefly (Kling, Seedance, Sora, Veo, and other video and image models), Adobe Photoshop, Adobe Premier.

   AI rating

   Speed

   2/ 5

   Brainstorm companion

   5/ 5

   Skills

   2/ 5

   AI credibility / professionalism

   1/ 5

AI is a great companion and speeds up some tasks enormously. Yet, at the moment, it requires more skills, not less. I used a whole stack of non-AI tools to make what I wanted out of the results AI gave me. Could I reach the same result with an AI-only set of tools? I believe so — but it would cost 100× more.