The Product Designer's New Stack Is Starting to Look Like a Software Team

A product design job used to have a stable tool requirement: know Figma, understand users, make prototypes, and communicate clearly with engineers. That description is becoming incomplete.

Recent remote openings at companies such as Automattic and Circle ask for something broader. Automattic wants designers who integrate AI tools, automation, and code-assisted workflows into daily practice. Its application form distinguishes simple content generation from coded prototypes and more advanced use of tools such as Claude Code, Cursor, and Lovable.

Circle is even more direct. Its design team uses AI tools every day to explore directions, prototype with code-generation systems, and stress-test ideas faster. The company lists Cursor, Claude Code, v0, and Lovable as useful examples of the workflow it wants candidates to demonstrate.

The signal is not that every designer must collect more software logos. The role is being reorganized around a new capability: turning uncertainty into something that can be experienced, tested, and shipped.

Figma is still the shared surface for visual design, interface systems, critique, and high-fidelity collaboration. But even Figma is moving beyond the static design file.

Figma Make is described as a prompt-to-app tool for turning ideas and existing Figma designs into functional prototypes, web apps, and interactive UI. Designers can attach frames and components, continue iterating through conversation, edit generated code, add a backend, and publish a working result.

That changes what a prototype can answer. A clickable sequence can communicate intent, but a functional prototype can expose validation problems, state changes, responsive behavior, awkward data, latency, and the moments where an AI system gives an unhelpful response.

The prototype stops being a polished story about a future product. It becomes a small piece of evidence about how the product might actually behave.

Cursor, Claude Code, and Codex matter because they work with the material of software rather than only with pictures of software.

Claude Code can read a codebase, edit files, run commands, build features, fix bugs, and work across multiple files. For a designer, that means asking where a component is defined, tracing how a design token reaches the interface, testing a real data state, or changing an interaction inside the product rather than describing the change in a handoff document.

This does not turn every designer into a senior engineer. It does reward designers who can reason structurally, inspect what the model changed, notice when an implementation is brittle, and collaborate with engineers before a concept hardens into a specification.

The handoff begins to shrink because design and implementation are no longer completely separate phases. Technical constraints can become part of exploration instead of arriving as a rejection after the presentation.

v0 and Lovable occupy a different part of the stack. They are useful when a team needs a new functioning product surface quickly, not only a change inside an existing repository.

v0 can generate real code and full-stack applications, create high-fidelity interfaces from wireframes or mockups, connect a backend, and deploy a live prototype. Lovable similarly generates editable applications with frontend, backend, database, authentication, and integrations from natural-language direction.

Their strongest use is not producing an entire company from one sentence. It is making comparison cheap. A team can test a dashboard against a conversational interface, try several onboarding structures, or place a proposed feature in front of users before allocating a full sprint.

AI prototyping lowers the cost of exploring an idea. It does not decide whether the idea deserves to exist.

The emerging stack increasingly includes React, Next.js, Tailwind CSS, component libraries, APIs, authentication, databases, Git, and GitHub. These are not merely engineering technologies placed beside design tools. They are becoming design materials.

A design engineer can work on the quality of an experience in both Figma and code. This is especially valuable for AI products, developer tools, complex dashboards, and interactive systems whose behavior is difficult to represent accurately in static frames.

Learning code no longer needs to mean leaving design. It can mean retaining authorship further into production, understanding the consequences of a decision, and making higher-quality tradeoffs with the people who will maintain the system.

The distinction that matters is not designer versus engineer. It is whether someone can move from an interface claim to a working interaction without losing the reason the interaction was designed in the first place.

The same transition is visible in media work. ElevenLabs asks its AI Creative Producer to prompt image, video, and audio models, produce cinematic and social content, document prompts, create repeatable workflows, stress-test products, and return structured feedback to the product team.

Superside describes a hybrid of generative video, filmmaking, cinematography, compositing, pacing, and editing. Its role still expects Premiere Pro and After Effects because generated material is not a finished story. It needs selection, continuity, timing, correction, sound, and a point of view.

A practical creative stack might move from Firefly or Midjourney for visual exploration, to Runway for generated motion, to ElevenLabs for voice, sound, or localization, and finally into Premiere Pro or After Effects for the work that makes separate assets feel intentional.

Generative models increase the amount of material available. Human direction determines whether that material becomes a film, a product demonstration, a campaign, or just more content-shaped noise.

Designers already choose materials according to their qualities. A public display behaves differently from a phone. A native app behaves differently from a responsive website. An AI model also has qualities, limitations, cost, speed, and characteristic failure modes.

Figma Make now exposes several model families through a selector. The interface remains stable while the underlying model can be changed during a conversation. That makes model choice visible inside an everyday design environment.

The same principle applies to generative media and voice. A faster model may suit a live interface. A more expressive model may suit narration. A video model may follow camera direction well but struggle to keep a character consistent. A stronger reasoning model may be worth its cost for a difficult product flow but wasteful for a quick layout variation.

Choosing the model is becoming part of the design decision, similar to choosing a framework, interaction pattern, production technique, or distribution channel.

Prompting matters, but prompt engineering is too narrow a description of what these jobs require.

The valuable practitioner can define the problem, provide useful context, distinguish a plausible result from a good result, connect generated work to an existing system, test it with users, identify technical and ethical risks, refine the output, and explain the process to the rest of the team.

AI makes it easier to generate options. It does not tell a team which option is coherent, trustworthy, accessible, commercially viable, or worth maintaining.

When production becomes cheaper, bad decisions can also be produced faster. Taste, judgment, research, systems thinking, and responsibility become more important because they are the filters between abundant output and a product people should actually use.

A portfolio for these roles should not be a gallery of AI-generated screens. It should show the movement from ambiguity to evidence.

For each project, explain what was unknown, what research shaped the direction, which tools and models were chosen, what functional artifact was created, what testing revealed, what changed because of that evidence, and what remained a human decision.

A rough working prototype can be more convincing than a perfect concept deck because it demonstrates that the designer can make the product behave. A documented failure can be useful when it shows that the designer recognized a limitation before shipping it to users.

The artifact is not proof that the model can generate. It is evidence that the practitioner can direct, evaluate, integrate, and take responsibility for what gets made.

The emerging toolkit includes Figma, Figma Make, Cursor, Claude Code, Codex, v0, Lovable, React, Next.js, Tailwind, Firefly, Runway, ElevenLabs, Premiere Pro, and After Effects.

But the stack is not the profession. The profession is still the work of understanding people, interpreting situations, making decisions, and giving form to systems.

What has changed is the distance between a decision and a testable result. Strong designers can now move through research, interface design, code, and media production without waiting for every discipline to hand the project to the next one.

That does not make one person a replacement for a software team. It makes them a more capable participant in one, with fewer excuses for keeping ideas abstract.