[2026 Latest] The Design Workflow Everyone Wanting to Delegate to AI Should Know First (With Ready-to-Use Templates)

The first thing people who want to create designs with AI get stuck on isn't choosing a tool.

It's "how much to leave to AI and where humans should start making judgments."

By reading this article:

Non-designers will learn how to request work from AI.

Directors will be able to make more concrete requests to designers.

Designers will find it easier to use AI as training wheels for ideation and verification, rather than seeing it as an enemy of production.

This article covers a practical AI design production workflow derived from cases at Dentsu Digital, Hakuhodo, CyberAgent, AIR Design, Goodpatch, Adobe, Figma, and Canva.

The header image was generated by AI. The prompt is

Published in an article over 80,000 characters Prompts are added daily.

Table of Contents

• Conclusion: The forefront of AI design has become a "Production OS"
• Common patterns seen in corporate case studies
• Tasks you should leave to AI vs. tasks you shouldn't
• 5-step AI design production workflow for practical use
• Ready-to-use request templates
• Common features of failed AI designs
• What to do today

Related Free Webinar

[End the AI Design Gacha.

Conclusion: The forefront of AI design has become a "Production OS"

The forefront of AI design has moved beyond the stage of simply "letting AI create images."

What is happening now is a movement to restructure the entire production process based on the premise of AI.

For example, Dentsu Digital's "∞AI Ads" supports ad creative production from "discovering appeal points" to "creative generation," "effectiveness prediction," and "improvement suggestions."

2026 Advertising and Marketing Trends in the AI Era as Envisioned by Japan's Top Ad Agencies

CyberAgent has introduced AI into all stages of ad production, from predicting ad effectiveness to design, copy, and AI talent.

Hakuhodo Products has established a new AI specialist role called "Generator" in their "AI Craft Studio," who uses multiple generative AI tools across the board.

In other words, winning companies don't just see AI as a "tool for making materials."

【The Future of Ad Creative Opened by AI】 Major Cases, Functions, Implementation Effects, and Outlook

They treat AI as a system that includes the pre- and post-production processes.

This is the key point.

What you should leave to AI is not making the final design right away.

What you should leave to AI are tasks like the following:

• Organizing competitors and reference cases
• Generating appeal ideas for different targets
• Generating rough design directions
• Creating variations of banners and thumbnails
• Creating drafts for LP (Landing Page) structures
• Generating image materials and background ideas
• Identifying improvement hypotheses
• Checking for omissions using checklists

Conversely, there are tasks that humans must finalize:

• Setting objectives
• Understanding customers
• Brand judgment
• Information architecture
• Readability of text
• Rights verification
• Final quality judgment
• Performance judgment after distribution

If you get this division wrong, you'll end up with a cheap design that just looks "AI-ish."

Common patterns seen in corporate case studies

Looking at domestic and international cases, companies that use AI well have commonalities.

It is about letting AI produce "quantity," humans deciding the "meaning," and using data to decide the "next improvement."

Dentsu Digital connects ad creative to ideation, generation, prediction, and improvement.

Hakuhodo i-studio developed an image generation AI system and internal workflow on the cloud to provide 10,000 types of visual content for LIFULL's SNS planning.

CyberAgent is moving toward mass-producing high-performance product images by combining effectiveness prediction with generative AI in "Goku Yosoku AI."

After introducing generative AI, the ad production team size shrank from "about 6 people of various roles" to "1 designer." Photography and talent costs were also reduced, and the number of productions per designer increased to about 170.

AIR Design uses AI and a professional team to analyze competitor LPs, banners, and search ads, connecting them to appeal ideas, design production, heatmap analysis, and improvement proposals.

Goodpatch is promoting the support for establishing AI-driven design and AI-driven development by incorporating AI into the design and development process.

Adobe Firefly is advancing on-brand image variation generation by providing Custom Models using corporate brand assets.

Figma is strengthening AI, design systems, code connections, and MCP integration.

Canva is moving toward a direction where even non-designers can easily produce without breaking the brand using Brand Kit and Magic Studio.

The commonality is clear.

AI does not become the "final decision-maker" in place of the designer.

AI's role is to increase the materials for judgment.

The flow companies are using is as follows:

1. Decide the purpose and customer
2. Generate a large number of ideas with AI
3. Humans select
4. Specialists finalize
5. Observe results through distribution and verification
6. Feed results back into the next production

Because of this loop, AI utilization changes from "playing around" to "business results."

Tasks you should leave to AI vs. tasks you shouldn't

The most important thing in AI design is separating the tasks you delegate.

Non-designers tend to skip this and just throw a request like "make it look good."

That will fail.

It's the same when requesting a designer.

You need to decide not just "please use AI," but "at which stage to use AI and which judgments will be made by humans."

Here is a guideline:

Tasks that are easy to leave to AI are those where there isn't just one correct answer and producing quantity is an advantage.

What humans should look at are judgments where responsibility is involved.

For example, it is effective to have AI generate 20 background ideas for a banner.

However, a human must check "whether this expression damages the brand," "whether there are issues with medical or advertising laws," and "whether customers will be misled."

Just having this line of demarcation significantly reduces AI design accidents.

5-step AI design production workflow for practical use

From here, we'll look at the order to use in actual work.

Decide the flow before the tool name.

The recommended 5 steps are as follows:

1. Decide the purpose in one line

The first thing you should write isn't a prompt.

It's the purpose.

Example:

For a beauty protein LP targeting women in their 30s, reduce anxiety about the first purchase and lead them to a subscription rather than a document request.

If this is vague, AI will only tidy up the appearance.

But it won't become a design that leads to results.

2. Fix the customer and medium

Even for the same product, the way you make it differs for Instagram ads, LPs, sales materials, and note header images.

Before requesting AI, decide at least these things:

• Who to show it to
• Where to show it
• How many seconds they have to judge
• What you want them to do next
• What impressions you absolutely want to avoid

If you start image generation without these five, you'll likely get lost.

3. Generate multiple directions with AI

This is where you use AI for the first time.

Don't aim for a finished product from the start; generate directions.

• Emphasis on trust
• Emphasis on price appeal
• Emphasis on expertise
• Emphasis on reviews
• Emphasis on Before/After
• Emphasis on worldview

By outputting different directions like this, you can compare them.

AI is better at "creating a state where you can choose" than "outputting one correct answer."

4. Human narrows down to one

The important thing here is not to choose based on preference.

Set the judgment criteria first.

• Does it resonate with the target?
• Is it readable on the medium?
• Does it align with the brand?
• Is there a difference from competitors?
• Does it lead naturally to the CTA?
• Is there no exaggeration or misunderstanding?

By looking through these criteria, even non-designers can make easier judgments.

5. Separate finishing and verification

Don't deliver what the AI output as-is.

A human must always finish it at the end.

• Font size
• Margins
• Color intensity
• Image glitches
• CTA visibility
• Smartphone display
• Rights and similarity
• Brand tone

Furthermore, if it's an ad or LP, look at the results after distribution.

The value of AI design isn't just in finishing the creation.

The value includes feeding the learning back into the next production.

**

Ready-to-use request templates

Whether you're throwing it to AI or requesting a designer, if the request text is sloppy, the output will be sloppy too.

Just using the following template will change things significantly.

1You are an art director for advertising creative.
2
3Please provide 5 design directions based on the following conditions.
4Do not make a finished design immediately; first, organize the "aim," "composition," "visual direction," and "expressions to avoid."
5
6【Purpose】
7{What you want to achieve}
8
9【Medium】
10{LP / Banner / note header image / Instagram ad / Sales materials, etc.}
11
12【Target】
13{Who will see it. Worries, knowledge level, anxiety before purchase}
14
15【Product/Service】
16{What to sell, convey, or have them apply for}
17
18【Action to take next】
19{Click / Application / Save / Inquiry / Document request}
20
21【Brand Tone】
22{Trust / Friendliness / Luxury / Speed / Expertise, etc.}
23
24【Information to include】
25{Product name, date, price, CTA, achievements, etc.}
26
27【Expressions to exclude】
28{Exaggeration, NG words, expressions to avoid for legal reasons, expressions too similar to competitors}
29
30【Output Format】
311. Direction Name
322. Aim
333. Layout Proposal
344. Main Copy Proposal
355. Visual Proposal
366. Points of Caution

The point is not to ask for a finished product.

What you ask for first are directions that can be compared.

After choosing a direction, request as follows:

1I will adopt Direction 3.
2Based on this direction, please concretize it into a banner proposal that is easy to see on a smartphone.
3
4Conditions:
5- Maximum of 3 text elements
6- Make the CTA stand out the most
7- Expressions that women in their 30s would find cheap are prohibited
8- If photo materials are needed, separate the background ideas to be made with generative AI from the conditions for searching for live-action materials
9- Finally, attach a quality checklist

It's the same when requesting a designer.

Instead of saying "This is a proposal made with AI. Please tidy it up," hand it over like this:

1Purpose:
2{Purpose}
3
4Adopted Direction:
5{Direction Name}
6
7Good points of the AI proposal:
8{Points to keep}
9
10Points of concern:
11{AI-ishness, text volume, brand inconsistency, rights, etc.}
12
13Points for the designer to judge:
14{Layout, color scheme, information priority, eye tracking, final tone}
15
16Delivery format:
17{Figma / Canva / PSD / AI / WebP, etc.}

This alone will raise the quality of the request.

The designer will also understand "what needs to be fixed."

Common features of failed AI designs

People who fail with AI design usually do the same things.

First is starting image generation without a purpose.

The appearance might be tidy, but it becomes a design that resonates with no one.

Second is deciding with only one proposal.

AI's strength is quantity.

It is suited for comparing multiple proposals rather than believing in just one.

Third is neglecting text.

In ads, LPs, thumbnails, and sales materials, the readability of text determines the results.

Even if the AI image is beautiful, it's meaningless if the text is unreadable.

Fourth is not looking at rights or similarity.

Even in the Ministry of Economy, Trade and Industry's "Guidebook for the Utilization of Generative AI for Content Production," considerations for intellectual property rights in content production using generative AI are organized.

If using it commercially, you must check the tool's terms of use, training data, similarity of the output, and the handling of people, trademarks, and characters.

Fifth is leaving "AI-ishness" as is.

Common AI-ishness includes:

• Margins are all the same
• Light is unnatural
• Hands or small objects are glitched
• Text is hard to read
• Faces are too perfect
• Too many decorations unrelated to the brand
• A cyber-feel seen somewhere before

The more you use AI, the more important human checking power becomes.

As AI utilization advances, the value of a designer will no longer be just "moving their hands."

What to keep, what to discard, and what to fix.

Value shifts here.

What to do today

Finally, I'll narrow down what to do today to one thing.

For your next design project, don't open an AI tool right away.

First, write these 7 items:

11. Purpose:
22. Target:
33. Medium:
44. Action to take next:
55. Information to include:
66. Expressions to avoid:
77. Judgment criteria:

Write this before requesting AI.

Don't look at the AI's proposal as a finished product.

Look at it as material for comparison.

Non-designers can stabilize their requests to AI by using these 7 items.

Directors can make their requests to designers more concrete by using these 7 items.

Designers can judge where to put AI in the production process by using these 7 items.

The essence of AI design is not to eliminate designers.

It's to reduce hesitation in production.

And it's to return time to the work that humans should judge.

This is the conclusion seen from the forefront of domestic and international cases.

Related Free Webinar

[End the AI Design Gacha.

Design from scratch with AI

NewsPicks Practice! Work Techniques

Contact us here

https://kawai-official.pages.dev/