Originally written: October 2025
When exploring the future of artistic workflows, we wrote this memo to formalize our thinking on why agents need specialized creative infrastructure. The core insight: AI will never replace human creativity—they need outlets for creative expression that complement their operators' processes.
Preface
Today, agents can generate images, edit videos, compose music, and design interfaces. But the second they need to integrate those creative outputs into a cohesive workflow with their human operators, the process breaks.
A designer has to manually wrangle agent outputs. An audio engineer has to export and re-import. A creative director has to stitch together disconnected artifacts.
That's a temporary state of the world. Within the next decade, agents will be seamless creative partners, not just tools. The infrastructure to make that happen doesn't yet exist. We're building it.
Specifically, we're creating outlets for creative expression that let agents work directly alongside human operators, across visual and audio media, starting with design and audio production workflows. Over time, we're consolidating the fragmented creative tools owned by Adobe, Autodesk, and DAW makers, making them agent-accessible from the ground up.
Entire creative industries are losing billions in velocity because agents can't participate naturally in creative workflows. Think design agents that can't iterate in real-time with human art directors, music agents that can't collaborate in shared audio sessions, or video agents that can't understand creative direction without extensive prompting.
We're giving them the creative outlets to do it, and once these exist, every agent-based creative workflow instantly becomes fluid and natural.
This is a once-in-a-generation moment. Creative tools have been effectively static for decades, and now they're about to be rebuilt, not for humans, but for human-agent collaboration.
Intro
The creative tool ecosystem has always been built around human-to-tool interaction. Photoshop, Premiere, Ableton, Figma—all designed for mouse and keyboard, for human timing, for human creative decision-making.
This works well today when humans are the sole creative actors. But as the world transitions into an agentic future, where agents are creative partners in visual and audio production, we need to create new ways for agents to express creativity that complements their operators' processes.
Agent-led creativity introduces many new questions: What creative formats do agents work best in? How do they understand and respond to creative direction? What's the feedback loop equivalent for iterative creative work? How do you keep creative tools low-latency so agents can participate in real-time creative sessions?
Today, agentic creative work makes up a tiny fraction of global creative output. But like the transition from desktop to cloud-based creative tools that Figma pioneered in 2016, the companies building agentic creative products today will, over the next decade, own a meaningful portion of all creative production globally.
These companies need better creative infrastructure built around their workflows to support the new opportunities that arise when agents become creative partners. Right now, there is an opportunity to re-invent creative tools to be faster, more collaborative, and more expressive for these human-agent teams.
Background
For simplicity's sake, we're going to consolidate creative work into two categories: visual media and audio media. Visual media refers to design, video, image creation, and motion graphics—think Figma, Premiere, After Effects. Audio media refers to music production, sound design, podcasting, and audio editing—think Ableton, Logic, Pro Tools.
These two categories have different characteristics and requirements, but both share similar underlying challenges when it comes to enabling agents as creative partners.
It's also important to define the types of agent creative interactions that do or will exist:
Agent-to-Human Creative Collaboration (A2H)
This is the primary creative pattern: an agent works alongside a human operator in a shared creative space. A designer gives creative direction to an agent, the agent generates options, the human refines, the agent iterates. In audio, a producer sets musical direction, the agent generates stems, the producer mixes, the agent adapts. This requires real-time feedback loops and shared creative context.
Agent-to-Agent Creative Collaboration (A2A)
Agent-to-agent creative work happens when specialized agents collaborate on the same creative project. A visual design agent and a motion graphics agent working together on an animated interface. A melody agent and a mixing agent working together on a track. While most focus is on A2H, A2A opens entirely new creative possibilities.
Agent-to-Output (A2O)
Sometimes agents need to produce creative outputs independently, then deliver them for human review. Marketing agents generating social media visuals, podcast editing agents producing final mixes, thumbnail creation agents. This requires consistent quality, brand alignment, and clear output specifications.
The Problems
Now that there's alignment on the creative categories and interaction patterns, let's examine the problems created in this future:
Creative Tools Break With Agent Collaboration
Traditional creative tools (Photoshop, Premiere, Ableton, Figma) are designed for interactions at human speed and with human interfaces. Using Photoshop to create an image involves clicking through menus, adjusting sliders, and seeing visual feedback. An agent can't meaningfully participate in this flow—it either generates something entirely independently, or the human has to manually integrate agent outputs.
Consider a design workflow: A human art director wants to explore logo variations. Today, they might prompt an image generation agent, download 10 options, manually import them into Figma, arrange them, get feedback, then start the process over. The agent has no concept of the broader design system, can't iterate in-place, and can't respond to "make it more playful" without starting from scratch.
When agents become true creative partners, and iteration cycles shrink from hours to seconds, it's imperative that creative tools let agents participate naturally in the creative flow.
Imagine a world where you're designing a website with an agent. You sketch a rough layout, the agent generates variations with proper design system tokens, you refine one, the agent adapts the entire system to match. This requires the agent to understand design context, iterate in shared space, and respond to abstract creative direction. Existing creative tools don't support any of this.
Creative Direction Is Unclear
Today, when a designer directs another designer, they use abstract creative language: "make it feel more premium," "give it more energy," "align with our brand voice." This works because humans share cultural context and creative intuition.
When agents become creative partners, translating abstract creative direction into actionable creative decisions becomes a new challenge. "Make it more playful" means different things for a children's book illustration vs. a fintech landing page. Agents need creative outlets that understand context, brand systems, and creative intent.
In audio, this is even more pronounced. "Give it more warmth" means different things to different producers, different genres, different emotional contexts. Music production agents need to understand musical context, genre conventions, and artistic intent to be effective creative partners.
Creative Context Is Lost
When humans create visually or audibly, they work within rich creative context: brand guidelines, style references, previous iterations, feedback from stakeholders, artistic direction. This context is largely implicit—stored in emails, Slack messages, mood boards, design system docs.
When agents participate in creative work, they need access to this creative context to make good decisions. A design agent generating social media graphics needs to understand brand voice, color systems, typography rules, and recent creative direction. A music production agent needs to understand genre preferences, reference tracks, mixing preferences, and artistic goals.
Today, this context is scattered across dozens of tools and human memory. For agents to be effective creative partners, creative context needs to be centralized, structured, and accessible.
Creative Tools Assume Human Timing
Creative software is built around human rhythm: the time it takes to move a mouse, the latency humans tolerate, the speed of human decision-making. When agents become creative partners, the timing changes dramatically.
An agent can explore 100 design variations in the time it takes a human to move a slider. An agent can generate and mix multiple musical stems faster than a human can listen to them. If creative tools maintain human timing constraints, they'll bottleneck agent creativity.
Future creative infrastructure needs to support both human and agent timing—slow enough for humans to follow creative evolution, fast enough for agents to explore creative space efficiently.
Visual Media Challenges
Design tools are built for pixel-perfect human manipulation: precise cursor movements, keyboard shortcuts, layer management. Agents don't "click" or "drag"—they need programmatic access to creative operations.
Current solutions (Photoshop APIs, Figma plugins) treat agents as external automation, not creative partners. An agent can batch-process images but can't iterate interactively with a human designer in real-time.
Video editing presents similar challenges. Agents can generate clips but can't participate in non-linear editing workflows, can't understand pacing and rhythm, can't respond to "make this section feel more urgent."
Audio Media Challenges
Digital Audio Workstations (DAWs) are built around human-paced timelines, manual MIDI input, and real-time audio monitoring. Agents can generate audio but can't collaborate in a shared DAW session, can't respond to "tighten up the drums," can't understand mixing decisions.
Music production is deeply iterative and context-dependent. A producer layers dozens of tracks, makes micro-adjustments to EQ and compression, and evaluates creative decisions by ear. Agents need outlets for musical expression that complement this process, not replace it.
Podcasting and voice work face similar challenges. An agent can transcribe and edit silence, but can't understand narrative pacing, emotional tone, or when a verbal stumble adds authenticity vs. needs cutting.
The Opportunities
It's clear that agents are making creative work more efficient. There's debate about creative displacement, but it's fair to say that many creative tasks done today by humans will be done collaboratively with agents.
The creative market is enormous. The global creative software market is $45B+ and growing. Design, video, and audio production represent hundreds of billions in annual creative output. As agents become creative partners, the companies building creative infrastructure for human-agent collaboration will capture meaningful portions of this value.
Creative Workspaces
The first opportunity is creating shared creative workspaces where humans and agents can work together naturally. These aren't traditional canvas tools and they aren't pure generative AI interfaces—they're hybrid creative environments.
For visual media, this means design spaces where agents can iterate alongside human designers, understanding design systems, responding to creative direction, and maintaining creative context across iterations.
For audio media, this means collaborative audio environments where agents can generate and manipulate audio while humans guide creative direction, mixing decisions, and artistic intent.
Creative Context Systems
Agents need access to creative context to make good decisions. This means building systems that capture, structure, and expose creative context: brand guidelines, style preferences, reference materials, previous iterations, stakeholder feedback.
This isn't just asset management—it's creative knowledge management. What makes this brand's visual identity distinct? What musical elements define this artist's sound? What creative principles guide this team's work?
Creative Direction Languages
Abstract creative direction ("make it more energetic," "give it warmth") needs to become actionable for agents. This requires new interfaces for creative direction—not just text prompts, but rich creative communication.
This might include visual references, audio examples, abstract creative parameters, or even direct creative manipulation that agents learn from. The key is maintaining human creative control while giving agents clear creative direction.
Iteration Frameworks
Creative work is inherently iterative. Agents need frameworks for creative iteration that match how humans work creatively: rapid exploration, refinement, version management, branching creative directions.
This means building creative version control, iteration tracking, creative option management, and collaborative review tools specifically for human-agent creative teams.
Real-Time Creative Collaboration
The most powerful creative moments happen in real-time: a design critique where a designer live-edits based on feedback, a recording session where musicians respond to each other, a video edit where pacing is felt not calculated.
Agents need to participate in these real-time creative moments. This requires low-latency creative operations, real-time creative feedback loops, and interfaces that support live human-agent creative collaboration.
Output Quality Systems
When agents produce creative work, quality consistency becomes critical. An agent-generated logo needs to match brand standards. An agent-mixed track needs to meet sonic quality expectations.
This requires quality frameworks: style consistency checking, brand compliance validation, technical quality assessment, and creative review workflows that ensure agent creative outputs meet standards.
Existing Players
The creative agent space is nascent, but several players are emerging:
Generative AI Tools (Midjourney, DALL-E, Stable Diffusion)
Focused on standalone creative generation. Strong at creating individual outputs, but no concept of creative collaboration, design systems, or iterative workflows. Treats agents as independent creators, not creative partners.
Creative Tool Plugin Ecosystems (Figma Plugins, DAW Scripts)
Attempting to bolt agent capabilities onto existing tools. Limited by legacy tool architectures. Can't enable true real-time collaboration or shared creative context.
Creative Automation Platforms
Building workflow automation for creative tasks. Good at repetitive creative production, but lack the flexibility and creative understanding needed for true human-agent creative partnership.
Distribution
There are three distribution strategies for creative agent infrastructure:
Creative Tool Integration
Start by integrating with existing creative tools (Figma, Adobe, Ableton) through plugins and APIs. Capture creators where they already work. This has limitations (legacy tool constraints) but provides immediate value and adoption.
Standalone Creative Workspaces
Build new creative environments designed for human-agent collaboration from the ground up. Higher friction to adopt, but enables capabilities impossible in legacy tools. The Figma strategy—steal market share by being fundamentally better.
Creative Agency Partnerships
Partner with creative agencies and studios to embed agent creative infrastructure in their workflows. Agencies are early adopters, have sophisticated needs, and influence broader market trends.
Risks
Two primary risks: creative agents don't mature fast enough to need specialized infrastructure, or existing creative tools adapt sufficiently.
Market Maturity
Today, most creative agents work independently—generate an image, produce a track. Collaborative creative workflows are rare. If this pattern persists, the need for human-agent creative infrastructure doesn't materialize.
However, we believe creative collaboration is inevitable. As agents become more capable, creators will want them as creative partners, not just output generators. The timing question is when, not if.
Incumbent Adaptation
Adobe, Autodesk, and DAW makers could add agent collaboration features to existing tools. If they do this well enough, new infrastructure may not be needed.
However, we believe legacy tool architectures fundamentally limit agent collaboration. Like Figma vs. Photoshop, sometimes you need to rebuild from scratch to enable new paradigms. True human-agent creative collaboration requires infrastructure designed for it from day one.
Conclusion
Creative agents need outlets for expression that complement their operators' processes. The infrastructure to enable natural human-agent creative collaboration doesn't exist yet.
We're building it—starting with visual and audio media, starting with creative workspaces that let agents and humans collaborate in real-time, starting with the creative context systems that let agents understand creative intent.
This is the beginning of a new era in creative work. Not agents replacing humans, but agents as creative partners—extending human creativity, accelerating creative output, and enabling creative possibilities that neither humans nor agents could achieve alone.
The creative tools of the future will be built for collaboration between humans and agents. That future starts now.
Additional Reading
- Figma. (2016). The Collaborative Design Tool - Adobe. (2024). Firefly and the Future of Creative Work - Ableton. (2024). AI in Music Production - OpenAI. (2024). DALL-E 3 and Creative Workflows - Anthropic. (2024). Claude and Creative Collaboration