Emerging AI design patterns. Part 2

Let’s continue. You can check the previous 10 patterns in Part 1.

Autocomplete

Autocomplete existed long before generative AI, but models’ advancement significantly expanded its capabilities. It’s no longer pattern matching but more a contextual understanding. And prediction can be based on any type of input including code, text, sketches etc.

Cursor

Canva

Memory

Usually an LLM’s knowledge about the user is limited to a single conversation. The memory feature enables AI to remember key details about the person across multiple chats and provides more personalized responses without repeating the same information. However, it’s essential to clearly inform users when their data is stored and have straightforward controls to review or delete it whenever desired.

Gemini

ChatGPT

Customise AI/Personality

Users can select specific traits of AI, such as behavior, interaction style, voice, tone, and others, so it better meets their needs and feels relevant for specific roles (like coach, personal assistant or nutrition expert). To make personalisation more accessible, it’s worth having a list of traits to select from in addition to free-form input.

Hume

ChatGPT

Create a custom voice/style

When creating content, it’s crucial to maintain consistency and adhere to brand guidelines or an individual’s manner of writing. AI products can learn to replicate the unique tone of voice by analysing examples of existing content or following specific voice features (like tone, formality etc). This approach extends beyond text to visual elements including photos, icons, and illustrations.

Claude

Grammarly

Krea

Canvas

Canvas is a specific interface that allows users and AI to collaborate on text or code beyond chat. It’s especially useful for projects that need many revisions. Unlike chat-based editing, which requires regenerating entire responses, canvas lets users directly edit content or highlight specific parts for AI modification.

ChatGPT

Gemini

Sources/Citations

Modern LLMs now can use external sources (such as internet or document collections) to generate a response and are no longer limited to data from model training alone. By providing a list of sources, AI products can clearly indicate where their answers came from and increase users’ trust in the results.

Perplexity

Google

Credits

AI computations are costly to run so companies often implement an internal credit system. It helps control expenses by charging users based on their actual usage of AI features and prevents potential abuse.

Krea

Visual Electric

Data management

AI companies collect user data to improve future models and enhance their services. While some platforms provide privacy settings to limit data sharing, not all allow users to fully opt out. Sometimes it’s available only for premium or enterprise users, while free users “pay” with their data to access the AI service, following the established business model of companies like Google or Facebook.

Figma

ChatGPT

Labeling AI-generated content

With the rapid improvement of AI models and the quality of their outputs, it’s becoming increasingly difficult for people to distinguish between human-created and AI-generated content. Clear labeling could work as an effective solution to address this challenge. This is particularly important for products where human interaction traditionally plays a central role, such as social media and video communications.

TikTok

Instagram

Agents

In 2025 AI agents will go mainstream and we expect to hear a lot about agentive UX (AX). What are AI agents? They are systems capable of understanding a task, formulating a plan, and executing it using advanced LLM capabilities such as reasoning, memory, vision and more. This is an extensive term that includes more patterns inside. Since the technology is still in its early stages, I’ll cover agent patterns in more detail once they mature.

Manus

Project Mariner by Google

Operator by OpenAI

The pace of AI development isn’t going to slow down anytime soon, so design patterns will continue to evolve and change. They play a key role in making AI approachable and enabling people to do amazing things with this technology.