Adobe, Figma and Microsoft vs. ArtCraft's AI-Built Clones: Which Software Moats a Rebuild Can Copy

Figma lost about half its market value in the first half of 2026, mostly on the fear that AI design tools would make it unnecessary. Then, at the end of September, a small team calling itself ArtCraft, led by Brandon Thomas, started publishing free rebuilds of Adobe's apps on GitHub, made largely by AI coding agents. Within three weeks it had a launcher listing thirteen apps, including FigCraft, a Figma clone, and WordCraft, GridCraft and DeckCraft for Microsoft Office. On October 13 HSBC cut Adobe to Hold and named ArtCraft as the reason.

I think ArtCraft is a useful piece of evidence, but not for the trade it seems to support. What the clones copied is the part of an app a stranger can observe from outside, and the maintainers' own notes show the gaps sitting exactly where the work stops being observable. Adobe's single-user creative seat lives on the observable side. Figma's moat mostly doesn't, and neither does Microsoft's. That makes HSBC's call on Adobe defensible and the market's pricing of Figma as the next casualty harder to defend.

Each Craft app also ships with a README that's unusually candid about how far it has to go. I've leaned on those more than on the launch coverage, because the maintainers have every reason to overstate and they mostly don't.

Two things stand out. The first is how much got built. A layered PSD that reopens correctly in a free app, a spreadsheet engine with nearly all of Excel's worksheet functions, and a word processor whose 422 commands an AI agent can call through MCP would each have been a funded company's multi-year roadmap not long ago. The second is where the gaps are. The work that's close to done is what you can watch from outside a product: the menus, the tools, the file format and how a document looks on screen. The work that's thin or absent sits off the installed app, in team accounts for FigCraft, co-authoring and VBA for the Office clones, and long-term signature validation for PdfCraft.

That split follows from how the apps were made. ArtCraft calls its method clean-room: an agent can observe what Photoshop or Word does and reproduce it, but isn't allowed to read their code. Observation gets you the interface and, with enough test files, the format. It doesn't get you a server-side system that thousands of other people are already using, because none of that shows up on one user's screen.

I've taken the four covered names with a Craft app aimed at them, plus PTC as a control, since no Craft app targets the product-lifecycle software where PTC's moat sits. The matrix shows the pillar a clone copies, the learned interface, next to four that a clone has to get past other people to reach.

Adobe is where the clone lands squarely. The framework had Adobe's learned interface weakened before ArtCraft existed, because generative tools were already lowering the skill a user needs, and ArtCraft's own pitch for PhotoCraft is "The open-source image editor you already know how to use." Adobe's bundle is weakened too, which matters for a free competitor. A Creative Cloud subscriber who only uses Photoshop is paying for a bundle they don't use, and that's the person most likely to try a free app that looks the same. Adobe's network pillar rests on PDF, and a clone that reads PDF doesn't change it, because PDF has been an open standard with free readers for years. What holds Adobe's document business is Acrobat and Sign as a record of contracts that have to validate later, which is the part PdfCraft's README lists as thin.

Figma's profile is the reverse. Its network effect is the strongest pillar it has, because a Figma file is where a designer, a few engineers and a product manager are all working at once, with the company's component library loaded alongside. FigCraft can open the file, and it ships a self-hostable collaboration server, but its README says there are no accounts, no teams and no cloud yet, and that its default public server stores files unencrypted. A design team can't move until the whole team moves, and an IT department won't sign off on that setup today.

Microsoft is the interesting case, because the test says the clone does reach one of its pillars. WordCraft covers 88% of Word's ribbon, so a single user's Office habit is now copyable, even though the framework treats Microsoft's interface as resilient on the grounds that Copilot works inside it. In a company, though, that habit is welded to the tenant: identity, Teams, SharePoint, compliance and the bundle they ship in. The Office READMEs don't mention co-authoring or VBA, so a company can't swap Excel for GridCraft without breaking the workbooks that feed each other through SharePoint and the macros its finance team relies on.

Autodesk and PTC fill in the edges. CADCraft reads and writes DWG, which matters for AutoCAD LT users who draft alone, but DWG has been readable by third-party tools for a long time, so it adds a cheaper option rather than breaking the partner network the framework credits. On my reading, the lock at Autodesk that matters most is Revit, the model a building is constructed from and shared with every partner on the project, and no Craft app goes near it. PTC is the control: its moat lives in Windchill's change history, which no outside observer can watch being used, and nobody has tried.

A clone doesn't need to win users to do damage. It can cost an incumbent in three ways, and they reach different names.

1. 1Lost seats. A user who can do the same work in a free app cancels. That needs the user to work alone, or the whole team to move together, so it falls hardest on single-user seats: Adobe's individual Creative Cloud plans, Microsoft's consumer plans, and to a lesser degree AutoCAD LT.
2. 2Capped price increases. A user who stays still knows there's a free alternative. That reaches anyone selling a seat, but it bites hardest where the product's own habit is the main reason to pay, which again is the creative desktop seat.
3. 3Procurement leverage. A large buyer cites a credible free option in a renewal. That only works once the option is credible to an IT department, which means accounts, admin controls, audited code and support, and ArtCraft offers none of those yet. According to Seeking Alpha's account of the note, HSBC told clients not to download the apps at all, citing security and maintenance risk.

Put together, the near-term exposure is Adobe's single-user creative seat. Adobe's results show why that matters and also why it isn't the whole company.

Neither company has reported a quarter that includes ArtCraft yet. Adobe's Q4 FY2026 is expected in December and Figma's Q3 2026 in November, and neither will say much, since alpha software doesn't move a quarter's revenue. Over the next year I'd watch whether Figma's retention holds up better than Adobe's creative segment growth. If the shared file is the moat the clone can't reach, it should.

The strongest argument against this piece is that I'm describing what ArtCraft hasn't built yet and treating it as something it can't build. Accounts, team workspaces, permissions and a sync server aren't exotic. They're ordinary software, much less novel than reproducing 27 Photoshop blend modes, and FigCraft already has a self-hostable collaboration server in its first public release. If a few people with AI agents got PhotoCraft from nothing to opening layered PSDs in a few weeks, they could ship hosted multiplayer with single sign-on in a few more. On that view the gap I'm relying on will close on a schedule.

The second argument is that agents make the learned interface irrelevant on both sides. Every Craft app exposes hundreds of commands through a command line, a JSON channel and an MCP server, so an AI agent can drive them directly. If more design and office work gets done by agents rather than by people clicking, the user's muscle memory stops being anyone's moat, and the question becomes which app is cheapest and easiest for an agent to drive. Free, local apps with a complete command surface fit that better than a subscription product built for humans. That would hurt Figma too, because Figma's push into Make, Slides and Sites courts product managers and marketers, the users with the least habit invested in the tool and the most willingness to let an agent pick the tool for them.

The third argument is that the clone doesn't have to win anyone over for prices to come down. A free app that looks like Photoshop gives every renewing subscriber, and every procurement team, a reference point. Microsoft's consumer plans have the same exposure, since a household using Word and Excel on its own doesn't need a tenant.

I think each of these is partly right, and I'd weight the third most, which is why Microsoft's consumer business is the one place I'd concede ground there. Where I disagree is on how much of a shared-file moat is the software. The hard part of moving a 200-person design organisation off Figma isn't the sync server. It's the component libraries the whole company builds on, the plugins the team depends on, the security review, the agencies and partners who send Figma links, and the risk that a migration breaks something nobody owns. FigCraft runs its own plugins rather than Figma's and hasn't been checked against real Figma documents, and those gaps need other people's cooperation to close, not just more code.

On agents, the mechanism points the other way from how it first looks. An agent working for a design team has to write into the file the rest of the team reviews, comments on and ships from, because work that lands in a local file nobody else opens hasn't been delivered. So the agent goes where the shared record is, and today that's Figma's cloud, which Figma already exposes to agents through its own MCP server. A clone wins the agent only once it also holds the team's shared file. If FigCraft ships accounts, teams and tested .fig import and teams still don't move, the counter-case was about code. If they do move, it was right and I'm wrong.

Adobe is the name the clone reaches, and HSBC's downgrade is defensible for the creative half of the business. Two things keep that from being the whole story. Acrobat and Sign sit on a system of record the framework rates intact, and PdfCraft's own README says long-term signature validation is the thin part. Firefly sits on licensed training data no clone can reproduce. A fair read is that ArtCraft adds pressure on the pillars Adobe was already losing rather than opening a new one.

Figma is where I think the market has it backwards. It has already been priced as the design name AI will commoditize, and a free clone of its editor looks like more of the same. The clone's own README says the opposite: the editor is the half it can copy, and the shared workspace is the half it hasn't built. FigCraft also had 11 GitHub stars when I checked on October 15, against about 61,000 for PhotoCraft, and Thomas has said he started the project to get away from Adobe's fees. Microsoft's enterprise business is barely touched on this evidence, though its consumer plans aren't. Autodesk's exposure is the AutoCAD LT desk rather than the Revit project. PTC, being bought by Schneider Electric, is outside the line of fire either way.

The lesson reaches past these five names. AI agents have made the visible half of an application cheap to copy, so a moat that rests mainly on users knowing the interface is now a pricing problem waiting to happen, and a moat that depends on colleagues, partners or counterparties being in the same record still has to be won one organisation at a time. That's the question I'll ask of every software name in coverage the next time a Craft app ships.