AI Is Reshaping Design: Hot Topics and a Forecast

For the past week I've been collecting AI-and-design hot topics automatically: seventeen sources a day, filtered, deduplicated and grouped into a briefing (see the full daily post). A single briefing is good for a quick scan, but when you line up a week's worth of material, a few clear threads emerge. This essay is a reworking of that material: AI, industrial design and open-source projects, ending with my own predictions.

AI × Industrial Design: three shifts in progress#

Hardware: AI devices finally have a product form worth discussing#

New details surfaced this week about the first device from Jony Ive and OpenAI: a screenless, donut-shaped speaker the size of a hockey puck, battery-powered, with a self-moving element that signals interaction states. It's the first consumer hardware from a top-tier design studio working with a leading AI company. Around the same time, Google open-sourced the Gemma Translator, a handheld translation device that runs the entire speech pipeline on a Raspberry Pi 5, with a 3D-printed, openly reproducible enclosure.

Both point in the same direction: AI hardware is moving from "a box with a screen" toward forms made for conversation. Without a screen, interaction has to be carried by form, material and motion — which is precisely industrial design's home turf.

Tools: the design toolchain is moving into the conversation#

Adobe shipped a unified ChatGPT plugin that brings Photoshop, Firefly, Premiere, Illustrator and 70+ tools into a conversational interface. Combined with Alibaba's Wan 3.0, which turns documents directly into video, the entry point of design workflows is shifting from software UI to natural language. We used to learn "how to use software"; now we learn "how to converse." Prompts are no longer optional — they're part of design execution.

Manufacturing: the distance between generation and physical object is shrinking#

Bambu Lab disclosed its 3D generative model research: reusing structural priors from a pretrained generative model to do part segmentation, beating previous best methods with only 0.32% of labeled data, and pointing toward automatic part-splitting and multi-material printing. This isn't an isolated event — the same week, open-source projects like Nurb, Kiln and CADAM are all working on natural-language modeling and agent-driven print preparation. Generating with AI is one thing; being able to print it is another, and the two are being connected.

Industry: the design process itself is being rebuilt#

At the "AI-powered brand going global" matchmaking event in Guangdong, manufacturers are already using AI to restructure innovation processes rather than merely assist drafting. NVIDIA's open-world models make "simulation as testing" a validation path for robot design. And Mixfont's decoy font reminds us that design must now address two kinds of readers at once: humans and AI. The industry's ask of designers is shifting from "can you draw it" to "do you know how AI participates, how to validate, and how to stay compliant."

AI project news: capability going mainstream, cost going down#

One thread runs through this week's project news: stronger, cheaper, more available.

• OpenAI opened unlimited text chat with GPT-5.6 Luna to free users, and gave Plus users a more fact-focused Sol;
• PPIO's Fusion model answers with multiple models in parallel and merges the results, reaching flagship-level quality at roughly a tenth of the cost;
• Ant Group open-sourced Avernet, multi-agent collaboration infrastructure tied to no single model;
• Google Maps' Ask Maps can now order food, book hotels and buy tickets — agents entering a billion-user product;
• Mirendil signed a $100M+ compute deal to pursue self-improving AI, a sign that frontier competition has shifted to infrastructure and long-horizon research.

For individual designers and small teams, the most direct change: AI capabilities that used to belong to big companies are entering everyday toolboxes at an affordable price.

Open-source projects: design tools going AI-native#

• OpenPencil: prompt-to-design-file vector tool with parallel agent teams — Design-as-Code;
• Onlook: billed as "Cursor for designers," visually building and editing React UIs while AI changes code with live preview;
• CADAM: a web app that generates parametric CAD models from a description;
• Nurb and Kiln: natural-language CAD modeling, and an MCP server connecting 3D printing to AI agents.

Taken together, design software is moving from "AI-assisted" to "AI-native with multi-agent collaboration." Open source lets anyone take these new forms apart and study them, which also means the next generation of design tools will iterate faster.

Predictions for the next year or two#

Based on this week's observations, here are the calls I feel most confident about:

1. The entry point of design workflows will migrate to conversation and agents. Software UIs won't disappear, but they'll recede; "tell the AI what you want, let it orchestrate the tools" becomes the dominant interaction.
2. Screenless conversational devices will become a hardware category, and industrial design — form, CMF, motion — will define their experience, the way it defined the smartphone era.
3. A personal "concept to object" loop will take shape: text-to-CAD plus agents plus 3D printing, shrinking the path from idea to prototype from weeks to days.
4. Design teams will become "one human plus a team of agents." Multi-agent orchestration will solve the who-does-what and how-do-they-align problems at the infrastructure layer.
5. Design will have two audiences. Humans and AI both "read" your work. Decoy fonts are only the beginning; adversarial information design becomes a real niche.
6. The cost of high-quality AI will keep falling. The capability gap between individuals and teams will be decided by workflow design and judgment, not budget.
7. Designers' new skill set will include three additions: collaborating with AI (prompts, agent orchestration), validating with simulation, and understanding AI content rights and compliance.

If I had to bet on just one, it's the first: the entry-point migration. Because it isn't a feature of any single product — Adobe, OpenAI and Alibaba all bet on it in the same month. When every big company moves in the same direction, it stops being a trend and becomes the terrain.

♪