Why a design tool for AI agents matters more to sellers than to designers
Every cross-border operator I know is running the same play on repeat: brief an AI tool to produce a listing image, a TikTok creative, or a landing page hero, watch it generate something that almost works, then spend an hour writing corrective prompts that get ignored by the next generation. The bottleneck was never the model’s ability to draw — it was the absence of persistent memory. Your brand guidelines, your past design decisions, your “we don’t use red because it converts poorly in Southeast Asia” notes — all of it evaporates the moment you close the chat window. That’s the real cost center hiding inside your creative operations, and it’s why a tool like Doop deserves your attention even though it won’t directly list a product or calculate landed cost. It’s attacking the coordination problem that sits between your creative team, your automation stack, and the AI agents you’re already paying for.
The actual problem: your AI agents have amnesia and no shared workspace
Here’s the scenario I see weekly in agency reviews. A brand running Amazon FBA and a parallel Shopify store decides to “do AI” for creative. They connect Claude to their workflow, ask for a set of A+ Content images, and get something decent. Then they ask for a matching TikTok Spark ad creative. The agent doesn’t remember the A+ Content decisions — the color palette, the typography hierarchy, the hero product angle — so it starts from zero. The output drifts. The brand’s visual identity fragments across marketplaces. And the human in the middle burns hours re-explaining context that should have been stored once.
Kevin Goedecke, the maker behind Doop, frames it as a design-tool problem, but it’s really an operations problem. He built Doop because most design tools treat AI agents like dumb render farms rather than collaborators with continuity. The pitch is straightforward: an infinite canvas where multiple AI agents work simultaneously, see each other’s cursors, stream designs frame by frame, and share one persistent memory of your decisions, taste, and context. The product positions itself as an open source Claude Design and Paper Design alternative, which tells you who they think the incumbents are.
The key architectural choice is that you bring your own AI. Doop doesn’t sell tokens or wrap models with markup. You connect the Claude or ChatGPT subscription you already pay for, and any MCP-compatible agent — Claude, Codex, Cursor, OpenClaw — can join the canvas. That’s a meaningful departure from the walled-garden approach most SaaS tools take, where they lock you into their model routing and their pricing per generation.
For a seller running a lean brand team, the appeal isn’t the canvas gimmick. It’s the memory layer. When you have one agent designing your Amazon listing images and another drafting your Shopify blog visuals, they’re not working from the same brief unless something forces them to. Doop’s approach — an editable markdown file for hard design rules plus a knowledge graph for softer preferences and decisions — is the closest thing I’ve seen to a shared brain for your creative agents.
How this actually differs from what you’re using now
Let me be specific about the incumbent landscape, because most sellers are already paying for tools that claim to solve this and demonstrably don’t.
Canva is where most sellers start. It’s fine for manual design, and its AI features generate passable assets. But Canva’s memory is you. You’re the one who remembers that the brand font is Montserrat, that the logo needs breathing room, and that the hero image for the German market shouldn’t show a hand gesture that reads wrong there. Canva doesn’t remember anything across sessions unless you build templates, and templates don’t capture why you made a choice — only the choice itself.
Figma is more powerful but has the same fundamental gap. It’s a multiplayer canvas for humans. The AI features are bolted on, not native collaborators. Your agents can’t join a Figma file, see what another agent is doing, and adjust in real time. And Figma’s version history is not a design memory — it’s a timestamped list of changes without the reasoning attached.
The AI-native tools like v0, Lovable, or even the design modes inside Claude and ChatGPT are generation engines. They’re great at producing a first draft and terrible at maintaining consistency across a campaign. Each prompt starts a new conversation with a fresh context window. You can paste your brand guidelines into every prompt, but agents don’t internalize them — they just follow instructions until the context gets long, then they drift.
Doop’s differentiator is the shared memory implemented as a knowledge graph, not just a text file. The maker’s response to a commenter’s question about where design memory lives is revealing: it’s stored in multiple ways, including an editable markdown file for hard rules (think DESIGN.md) and a knowledge graph under the hood that captures design decisions and style preferences. They use Cognee for the memory database. That’s a meaningful implementation detail — it means the system can retrieve relevant design context based on similarity, not just keyword matches.
For comparison, the closest incumbent is probably Framer or Webflow on the website side, but neither has a multi-agent collaboration model. And on the pure AI side, tools like Midjourney or DALL-E have no memory at all — every generation is a roll of the dice unless you meticulously engineer prompts.
The practical difference for a seller: with Doop, you can have one agent working on your Amazon product image stack while another handles your TikTok Spark ads, and both pull from the same design memory. When you decide that the lifestyle shot should use a warm filter for the US market but a cooler, more minimal aesthetic for the EU, that decision is stored once and both agents respect it going forward. That’s not a feature — that’s a workflow change.
Why Amazon sellers should care more than Shopify ones
I’ll make a contrarian claim: this tool matters more if you’re an Amazon FBA operator than if you’re running a DTC Shopify brand. Here’s the reasoning.
Shopify brands control their entire visual ecosystem. They own the theme, the product pages, the email templates, the social assets. If they want consistency, they can enforce it through their theme files and template libraries. The AI agents they use for creative are tangential — nice to have, but not load-bearing.
Amazon sellers don’t have that luxury. Your listing images, A+ Content, and brand store are constrained by Amazon’s template system. You’re uploading flat images that have to follow specific dimension and content rules. The creative decisions — which lifestyle image converts, which infographic style reduces returns, which angle highlights the product’s differentiator — are made in a fragmented way across your team and your AI tools. And once you’re selling across multiple marketplaces — US, UK, DE, JP — the same image set needs localized variations. The person or agent producing the German version needs to know why the US version used a particular callout box, and whether that reasoning still applies.
Amazon’s own tools, like Brand Story or the A+ Content manager, don’t have any AI memory. You’re uploading assets, not collaborating with an intelligent system. Doop, or something like it, gives you a layer above Amazon’s rigid templates where your agents can work out the variations before you export the final files and upload them to Seller Central.
The other angle: Amazon sellers typically run thinner creative teams than DTC brands. A solo operator or a two-person team managing multiple SKUs and marketplaces can’t afford to re-explain brand context to an AI tool every time they need a new creative. The memory layer isn’t a luxury — it’s the difference between AI creative being viable and being a time sink.
What cross-border sellers can actually borrow from this
You don’t need to adopt Doop specifically to benefit from the pattern it represents. But you should steal the underlying architecture for your own creative operations. Here’s the breakdown.
The DESIGN.md pattern. The idea of a living markdown file that stores hard design rules is transferable to any team, whether or not you use an AI canvas. Write down your non-negotiables: brand colors with hex codes, typography stack, logo usage rules, image style preferences per marketplace, cultural do-nots for each region you sell into. Store it in your repo, your Google Drive, your Notion — wherever your team actually works. Then every AI tool you use gets pointed at that file as the first context. This alone will reduce creative drift more than any tool upgrade.
The knowledge graph for softer decisions. Hard rules are easy. The harder part is capturing the why behind decisions. Why did you switch from a white background to a lifestyle background for the hero image? Why does the Japanese listing use a different callout structure than the US one? Those decisions live in Slack threads and meeting notes, not in any design file. Doop’s approach of storing these as graph nodes with relationships is worth copying conceptually. Even a simple structured document — decision, context, date, owner — will improve how your team and your AI tools handle future creative briefs.
Multi-agent review workflows. The most interesting part of Doop’s pitch isn’t the canvas — it’s the idea that agents review each other’s work. One agent designs, another critiques, a third checks against the brand memory. For sellers, this maps to a quality-control layer you probably don’t have. Before an AI-generated creative goes live, have a second agent check it against your design rules and marketplace requirements. This catches the classic failures: wrong image dimensions for Amazon, text that gets cut off on a mobile TikTok view, a color that violates a platform’s guidelines.
The bring-your-own-AI principle. Doop’s choice to let you connect your existing Claude or ChatGPT subscription rather than forcing you into their token ecosystem is a lesson for how you should evaluate AI tools generally. If a tool makes you pay for AI usage on top of your existing subscription, the pricing math usually breaks. You’re paying twice for the same model. Tools that integrate with your existing subscriptions are structurally more aligned with your cost base.
Where the math breaks
Let me be direct about the limitations, because the Product Hunt launch energy doesn’t tell the whole story.
First, this is an early-stage tool. The Product Hunt page shows a launch with four comments and a handful of reactions. There’s no mention of pricing, no enterprise tier, no case studies with real brands. The maker is responsive in the comments, which is good, but there’s no evidence of production-scale usage. For a seller running a serious operation, adopting a tool this early is a bet, not a decision.
Second, the multi-agent canvas model assumes you have multiple AI agents working on creative simultaneously. Most cross-border sellers don’t. A typical operation might use one AI tool for image generation and occasionally experiment with another for copy. The workflow Doop enables — multiple agents collaborating, reviewing each other, sharing memory — requires a level of AI integration most teams haven’t reached yet. This is a tool for the leading edge, not the mainstream.
Third, the memory layer is only as good as the underlying model’s ability to use it. The knowledge graph stores information, but the agents still have to retrieve and apply it correctly. Anyone who’s worked with AI agents knows they drift, especially in long sessions. The DESIGN.md file helps, but it’s not a guarantee. You’re still going to catch errors in the output, and you’re still going to need a human in the loop for final review.
Fourth, there’s no mention of how this handles the actual output formats sellers need. Can it generate Amazon-compliant image stacks with the right dimensions and safe zones? Can it produce TikTok-native vertical formats? The launch page talks about designing ad creatives — one commenter mentions using it for that purpose — but there’s no detail on export presets or platform-specific templates. That’s a gap for sellers who need production-ready files, not just beautiful canvases.
Finally, the open source angle cuts both ways. Doop is open source, which is great for transparency and customization. But it also means you’re responsible for maintaining your own instance if you go that route. The hosted version might be more convenient, but the launch page doesn’t clarify what’s hosted versus self-managed. For a non-technical seller, that ambiguity is a barrier.
The bigger shift: from prompt engineering to memory engineering
Stepping back from Doop specifically, the launch signals something larger for cross-border e-commerce operations. The competitive advantage in AI-assisted creative is shifting from who writes better prompts to who builds better memory systems.
Prompt engineering was the first wave — learning to phrase requests so the model produces usable output. That’s table stakes now. The second wave is context engineering — building systems that store, retrieve, and apply your brand’s knowledge across multiple AI interactions. Doop is an early example of this, but the pattern will spread to every AI tool you use.
For sellers, this means your creative operations should be designed around memory, not around individual generations. Every AI tool you use should have access to your brand rules, your past decisions, and your marketplace-specific requirements. The tools that support this — whether it’s Doop or something else — will compound in value because every generation gets better as the memory grows. Tools that don’t support this will feel increasingly frustrating as you realize you’re re-explaining context you already paid to establish.
The practical implication: start building your brand memory now, regardless of which tool you use. Write the DESIGN.md file. Document your decisions with reasoning. Structure your creative assets so that context is attached to output. When the next generation of AI tools arrives — and it will, quickly — you’ll have the memory infrastructure ready to plug in.
What I’d watch / test next
If you’re intrigued by the pattern but not ready to commit to an early-stage tool, here’s what I’d do this week.
First, take one SKU or one campaign and write a DESIGN.md file for it. Include the hard rules — colors, fonts, image styles, marketplace-specific requirements. This takes an hour and pays off immediately, even if you never touch Doop.
Second, test Doop with a low-stakes creative task. Connect your existing Claude or ChatGPT subscription via MCP, put two agents on a canvas, and ask them to produce a set of ad creatives for one product. Evaluate not just the output quality but the collaboration — do the agents actually respect the shared memory? Do they catch each other’s mistakes? The comment from Tim Osterbuhr at Poof mentions using Doop for ad creatives with Claude, which is a reasonable starting point. The installation is straightforward — desktop or web app, then hook up your subscription via MCP, as the maker explains in the comment thread.
Third, monitor how the memory layer evolves. The maker’s response about using an editable markdown file plus a knowledge graph with Cognee is worth watching. If that memory layer becomes robust enough to handle real brand complexity — multiple product lines, multiple marketplaces, years of design decisions — it becomes genuinely valuable infrastructure. If it stays at the level of “design preferences,” it’s a nice-to-have that won’t move your metrics.
Fourth, check whether the tool develops platform-specific export presets. The moment Doop can output Amazon-compliant image stacks or TikTok-native formats directly, it becomes materially more useful for sellers. Until then, it’s a design tool that produces assets you’ll still need to process for each platform.
The bottom line: Doop is worth a look, not because it’s the finished solution for cross-border creative operations, but because it’s an early signal of where AI tools are heading. The tools that win your workflow will be the ones that remember what you’ve decided — not the ones that generate the prettiest first draft. Start building that memory now, and you’ll be ready when the category matures.
