How I Stopped Doing Routine Work Manually: AI Tools We Use at a Design Studio

WhisperFlow — when you no longer need to print

I’ll start with the most unobvious. I almost stopped typing — neither on the computer nor on the phone. Instead I just say it. WhisperFlow is AI-powered voice input. You press a key combination, speak, and the text appears in any field, in any application. The difference from standard voice recognition is that a neural network works here: it corrects errors, rearranges words for better readability, and understands the context. Especially convenient in English. My English is not perfect, but WhisperFlow seems to interpret what was said — and the output is a normal text. The developers are talking about acceleration by five times, I would say exactly two times. But even this changes everything. When I created the landing page using vibe-coding, I literally didn’t write a single line of text by hand. I just dictated.

Why I chose Claude

My path was standard: first ChatGPT, then Gemini, now Claude. I liked each subsequent instrument more than the previous one — and so far I have settled on Claude. I bought a subscription for $100 per month. It sounds expensive, but I use it for literally everything: work tasks, personal issues, project evaluation, writing, coding. It pays for itself in time — that’s for sure. The main thing I like is the minimal error rate. I use the most powerful model, it spends tokens quickly, but the result is predictable. When you work through Projects and upload skills there, the necessary context is always at hand, you don’t have to explain everything from scratch every time.

Estimating projects in minutes, not hours

This is perhaps the most valuable case for me as a studio manager. I made a skill — a custom prompt inside Claude Projects — specifically for evaluating incoming projects. Our business process, client response template, and the logic for rounding hours and costs are embedded there. Now when a request comes in, I throw materials from the customer and one text into Claude. Then he himself: translates the technical specification if it is in English, formulates clarifying questions for the client, makes a detailed estimate in hours by stages, calculates the range in days and money, and writes a ready answer to the client. The answer turns out so good that I hardly edit it — sometimes I’ll add a couple of my examples, and that’s it. If the client is targeted, I spend more time checking. But the starting point is already ready. The skill is very simple to do — Claude himself guides you step by step, you simply answer the questions.

Presentations and CP — from brief to file

I load the necessary graphics into Claude, dictate through WhisperFlow what I want for each block — and it assembles PowerPoint. I then finish off some points, but the main work is done. For the factory, Svoboda, for example, asked him to make it in our style — he even took a hat from our presentation. But something else is more interesting — landing pages as KP.

Instead of the usual presentation, Claude can put together a responsive HTML landing page in a few minutes. It looks professional, opens on a mobile phone, you can add a metric and see where the client pays attention. Hosted on Vercel — free, fast.

Claude in the browser — AI that runs in the background

There is a Claude plugin for Chrome that allows you to literally tell Claude what to do on the site — and he goes and does it. Just like a person: clicks, looks, analyzes. I used this to analyze the Svoboda factory website — a large site, manually browsing through the pages would take a long time. I wrote to him: go through the entire site and look at the types of pages. He started walking and looking, I left it in the background and went about other things. Then I checked — it was distributed correctly, with a couple of small adjustments. I also used it to fill out tables — so as not to transfer data by hand. This is boring work like Ctrl+C, Ctrl+V, which is easy to delegate to the background. The only negative is that it is still slow and wastes a lot of tokens. For predictable, repetitive tasks, it is better to write scripts — faster and cheaper. But for tasks with an element of analysis, it’s just right.

A script that sends messages itself

We have accumulated a database of applicants — almost 800 people. We needed to send them information about our referral program. Doing this manually takes several days of work. I asked Claude to write a Python script to send automatically via Telegram. He wrote and took into account the necessary nuances — in particular, that Telegram blocks too fast automatic sending. I run the script in the background and go about my business. In four weeks the entire base will be covered. I don’t write code myself — but with Claude you don’t have to. You describe the problem, he writes the solution.