The design tool is no longer the destination — it’s the interface your agent writes to
Every social media operator I know has quietly become a designer. Not by training, but by necessity: the carousel, the thumbnail, the Story frame, the LinkedIn document, the Pinterest pin. The work that used to live with a design team now sits in a Canva tab next to a scheduler, and the bottleneck isn’t taste — it’s the mechanical translation layer between an idea and a file that’s correctly sized, on-brand, and exportable. That’s why a launch like Tidy matters to this audience even though it isn’t a social tool. Tidy is an early-stage, open-source, self-hostable alternative to Figma and Paper aimed at “designers, engineers, and AI agents” working together, and its core bet is that coding agents — Claude Code, Codex, OpenCode, and others — should be able to create and edit designs directly inside the design tool via MCP. If that bet lands, the way small content teams produce visual assets changes shape: you stop clicking through a canvas and start describing outcomes to an agent that manipulates the file. My take: this is the most interesting thing to happen to the creator toolchain in months, and it’s also nowhere near ready for a social team’s production pipeline. Both things are true.
The actual problem Tidy is pointing at
Strip away the launch-page framing and Tidy is making a claim about where design work goes next. The maker, Tadhg Dowdall, describes building it with his brother Sean out of Ireland, framing it as an “alternative to Figma and Paper for designers, engineers, and AI agents to work together.” The mechanism is MCP — the Model Context Protocol — which lets a coding agent like Claude Code or OpenAI’s Codex read and write into Tidy’s document model rather than just emitting code you paste somewhere else.
That distinction is the whole story. Right now, if I ask an agent to “make me a 1080x1350 carousel slide with our brand tokens,” what I get back is either a React component I have to render, an SVG I have to eyeball, or a Python script that draws rectangles with Pillow and looks like it. The agent can generate the artifact but it can’t see the artifact, and it certainly can’t iterate on it the way a designer does — nudging spacing, rebalancing type, checking whether the headline actually fits. An agent that writes into a real design document with a real object model closes that loop. That’s the theory, anyway.
For social operators, the practical version of this problem shows up weekly. You have a campaign concept, you need it in five aspect ratios across four platforms, and the variation work is pure drudgery. The reason tools like Canva won — and why CapCut ate short-form video editing — is that they collapsed the distance between idea and export. Tidy is betting the next collapse comes from agents doing the variation work inside a structured document instead of a template picker.
Why this is a repurposing story before it’s a design story
I’ve written before about how repurposing is the highest-leverage activity in a small content team, and the bottleneck is almost never the writing — it’s the reformatting. A single pillar piece becomes a LinkedIn carousel, an Instagram Reel cover, a YouTube thumbnail, three X image cards, and a Pinterest pin, and each one is a slightly different design problem. In my own tests of agent-assisted design workflows, the failure mode is always the same: the agent produces something plausible at one size and falls apart at another because it has no persistent model of the layout. A tool that gives the agent a real document to edit — with layers, constraints, and objects — is exactly the missing piece. Whether Tidy is that tool is a separate question from whether the category is right, and I think the category is right.
How it differs from what you’re probably already using
The honest comparison set here isn’t really Figma, because almost nobody running social accounts is laying out carousels in Figma by hand anymore. The real incumbents are Canva, Adobe Express, and whatever template library your scheduler ships with. Those tools are template-first: you pick a shape, you swap the text, you export. They’re fast, they’re collaborative, and they’re deliberately shallow — which is a feature, not a bug, for a two-person marketing team.
Tidy is document-first and agent-first. The pitch is that a coding agent connected over MCP can create and edit designs in the tool itself, which means the workflow looks less like “open template, edit text” and more like “describe the asset system, let the agent build and revise it, review the output.” That’s a fundamentally different interaction model, and it’s closer to how engineering teams already work than how marketing teams do.
The other differentiator is licensing. Tidy is open-source under Apache 2.0 and self-hostable, per the maker’s own description. That matters more than it sounds. Self-hosting means a team can run it on its own infrastructure, which is the kind of thing that gets a tool past legal review at a company that won’t put unreleased campaign assets in a third-party SaaS. It also means the tool can be extended — if your brand has a weird export requirement, someone on your team can write it. Compare that to the closed ecosystems of Canva or Adobe, where you get the integrations they ship and nothing else.
There’s a third comparison worth naming, which is the emerging class of “agent-native” creative tools — things like v0 for UI or the various image-gen endpoints wired into schedulers. Most of those generate assets outside a persistent design document. Tidy’s bet is that the document is the durable artifact and the agent is the operator. I’d bet that’s the right architecture, because it’s the only one where revisions compound instead of resetting.
What creators and social teams can actually borrow from this
Even if you never install Tidy, the launch is a useful signal about where your workflow should be heading. Three things stand out.
Build the brand system, not the asset
The reason agent-driven design is even plausible is that Tidy assumes a structured document — objects, layers, a real model — rather than a flat canvas. The lesson for social teams is that the highest-value thing you can do this quarter is codify your visual system: exact type scales, color tokens, spacing rules, logo clear-space, safe zones per platform. If your brand lives in a PDF style guide nobody opens, an agent can’t use it. If it lives as tokens and constraints, an agent can. This is unglamorous work and it pays off regardless of which tool wins.
Treat MCP as the integration layer to watch
The Model Context Protocol is becoming the default way agents talk to software, and Tidy’s decision to connect Claude Code, Codex, and OpenCode through it is the most forward-looking thing on the page. For operators, the practical implication is that “does this tool have an MCP server?” is about to become a real procurement question, the way “does it have an API?” was a decade ago. When I evaluate scheduling and design tools now, I’ve started asking whether they expose a machine-readable interface, because the tools that do will be the ones that survive the agent transition.
Self-hosting as a content-ops hedge
The Apache 2.0 license and self-hostable claim are worth flagging for teams with compliance constraints. Not every creator needs this. But if you’re at a company where marketing can’t upload unreleased product shots to a random SaaS, an open-source design tool that runs on your own infrastructure is a genuinely different risk profile. It’s also a hedge against pricing changes — the thing that has burned every social team that built its workflow on a tool that later repriced.
Why short-form video creators should care less than carousel people
If your output is Reels and TikToks, Tidy is not your problem — your bottleneck is editing, and CapCut and its competitors own that. The people who should care most are the ones producing static, layout-heavy formats: LinkedIn document carousels, Instagram carousels, Pinterest pins, YouTube thumbnails, X image cards. Those are the formats where a structured design document plus an agent is a real unlock, because the work is fundamentally compositional rather than temporal. I’d bet the first wave of real adoption lands with B2B content teams and newsletter operators who live in carousels.
Where my judgment says it falls short
Here’s the part the launch page doesn’t say, and I’m flagging it as opinion because the source is silent on all of it.
It’s very early. The maker says so directly — “we’re very early on” — and that’s the single most important fact on the page. Early-stage design tools are missing the boring things that make them usable: reliable export at platform-specific dimensions, font handling, asset libraries, comment and review flows, version history that a non-engineer can navigate. None of that is mentioned, and I wouldn’t assume any of it exists.
The audience is developers, not marketers. Read the description again: it’s aimed at “designers, engineers, and AI agents.” The example agents are coding agents. The interaction model assumes you’re comfortable connecting an MCP server and prompting a CLI. That is not the median social media manager, and it’s not close. If you’re a content operator who has never touched a terminal, this is not a tool you can adopt today — it’s a tool you should watch.
No pricing, no user numbers, no roadmap. The source doesn’t disclose pricing, user counts, or a public roadmap. That’s normal for an early open-source launch, but it means any claim about maturity is speculation. Don’t let anyone tell you otherwise.
The “alternative to Figma” framing oversells the comparison. Figma is a mature multiplayer design platform with a decade of collaboration infrastructure. Tidy is an early open-source project. The comparison is directional, not literal, and treating it as a head-to-head would set the wrong expectations.
Agent reliability is the open question. Even with a perfect MCP integration, agents are inconsistent at aesthetic judgment. In my experience testing agent-assisted layout, the agent gets you 70% there and the last 30% — the part that makes a carousel actually look good — still needs a human eye. Tidy doesn’t claim to solve that, and nobody should expect it to.
What I’d watch / test next
This week, three concrete things. First, if you run a content team, spend an hour documenting your visual system as tokens and constraints — type scale, colors, spacing, per-platform safe zones. That’s useful no matter what tool you use, and it’s the prerequisite for any agent workflow. Second, if you’re technically inclined, spin up Tidy from its open-source repo and try one real task: have Claude Code or Codex generate a five-slide carousel inside it, then judge the output honestly against what Canva would have produced in the same time. Third, add “does it have an MCP server or a machine-readable API?” to your tool evaluation checklist alongside pricing and integrations, because that’s the axis the next two years of tooling will be sorted on.
What I’m watching for from Tidy specifically: an export pipeline that handles platform dimensions without manual intervention, any sign of a non-developer onboarding path, and whether the open-source community builds the boring collaboration features the core team hasn’t. If those three things show up, this becomes a real option for social teams. If they don’t, it stays a fascinating proof of concept — and the lesson about where design work is heading still holds.
