Between 30 September and 15 October 2026, one small company and its AI agents published open-source copies of thirteen flagship desktop apps — Photoshop, Word, Excel, PowerPoint, Acrobat, Premiere, Illustrator, InDesign, After Effects, Lightroom, AutoCAD, Pro Tools and Figma. Free, in one programming language, all working, all early. It's the clearest sign yet that copying what software looks like has become cheap. What it doesn't copy — the part below the waterline — is where your software bill actually lives.
The company is ArtCraft (GitHub: storytold), a US start-up whose paid product is a desktop tool for making AI images and video. Its founder previously built the AI voice sites FakeYou and Storyteller.ai. Three people are public members of its GitHub organisation. The “Craft” apps are given away under permissive open-source licences, and every one of them points back to ArtCraft.
| App | Modelled on | Commits | Version |
|---|---|---|---|
| PhotoCraft | Adobe Photoshop | 2,371 | 0.9 |
| LightCraft | Adobe Lightroom | 2,895 | 0.8 |
| VectorCraft | Adobe Illustrator | 2,848 | 0.11 |
| FilmCraft | Adobe Premiere | 1,951 | 0.8 |
| EffectCraft | Adobe After Effects | 1,875 | 0.10 |
| PdfCraft | Adobe Acrobat | 1,165 | 0.8 |
| DesignCraft | Adobe InDesign | 1,025 | 0.8 |
| GridCraft | Microsoft Excel | 1,015 | 0.7 |
| DeckCraft | Microsoft PowerPoint | 884 | 0.7 |
| WordCraft | Microsoft Word | 727 | 0.7 |
| SoundCraft | Avid Pro Tools | 682 | 0.7 |
| CADCraft | Autodesk AutoCAD | 488 | 0.7 |
| FigCraft | Figma | 270 | 0.6 |
Measured from GitHub on 15 October 2026. Line count estimated from GitHub's per-language byte totals, calibrated against a direct count of WordCraft (~240k lines). A fourteenth app, AniCraft, was announced on 14 October with no code yet.
This is not thirteen teams. It's one production line: every app shares the same instruction file for the AI agents, the same quality rules, the same build checks and the same layout. People decide what to build and police the rules; agents write the code. The rules are what turn hundreds of agent commits a day into software instead of mush.
Look at what the projects say about themselves. WordCraft has 97% of Word's ribbon and menu commands wired up — and rates itself only ~60% “ready for real work”, with zero users reporting they've switched from Word. Its own number-one gap: it has never been tested against a body of real documents from Word, Google Docs or LibreOffice. That gap between the two numbers is the whole story.
Menus, buttons, the familiar layout. This used to take a team years; agents now produce a convincing version in weeks. It's what wins the screenshot and the star count.
The supplier's odd spreadsheet that has to open correctly. The tender document that must print on exactly 40 pages. The macro finance built in 2014. The person you call when it breaks. None of that is in the menus.
It's the same shape as your own business: the front office customers see is the small part; the back office is where the cost, risk and knowledge sit. Software vendors are no different. Copying their front office got cheap. Their back office didn't.
Deeper: The iceberg →“There is no alternative” is weaker every month. You don't need to switch to use that — a credible free option for the light users is leverage when the per-seat price goes up again.
Most organisations pay flagship prices for people who open, read and lightly edit. That's the slice free tools reach first. Count how many of your seats are really heavy users.
Every Craft app can be driven by an AI agent without a screen. Generating and editing Word, PDF or layered image files on your own servers, with no per-seat licence, is usable today — well before the apps are ready for people.
If a small team can stand up a Photoshop-shaped editor in ten days, your internal tools, portals and that niche package you overpay for are all cheaper to replace than your last estimate assumed.
Don't swap everything at once. Move one group of users or one job at a time, keep the old licence as a safety net, and only cut seats when the new tool has handled real work for a quarter. Cheap copies make that path cheaper; they don't make the big-bang switch any safer.
Deeper: Strangle the licence →Every Craft app is version 0.x, releasing almost daily. The makers are unusually honest about it: PhotoCraft published that 48 of its 146 settings do nothing and that its tests “miss what users hit”. WordCraft lists open crash reports on common hardware. Agents shrink the time it takes to build a plausible feature. They don't shrink the time it takes real users to find out whether it works.
Legal attention. Naming Microsoft, Adobe, Autodesk and Avid products as the thing you're copying invites a response. The projects follow strict “clean-room” rules — observe the original, copy nothing — and copying menus and file formats has strong legal precedent. But names and look-alike designs are where disputes start. Not legal advice; watch for renames.
A tiny team behind fourteen products. Over 5,000 open issues and 3,700 open change requests across the thirteen apps, two weeks in. Keeping that healthy after the launch excitement fades is the open question.
Stars aren't users. 150,000 stars measure a viral launch. Even long-dead repositories in the same account suddenly picked up dozens of copies in the same week — a sign of attention, some of it automated, not adoption.
Free isn't the business model. The apps are a funnel to ArtCraft's paid AI product. That's legitimate, but it tells you where the company's priorities will go when the two compete for time.
For SA organisations, Adobe and Microsoft seats are priced in dollars and move with the rand. Schools, NGOs, small agencies and municipalities pay the same per seat as a New York firm. Free, offline, permissively licensed tools that also run in a browser are worth tracking closely — but as a lever and a pilot, not a migration. Documents that go back and forth with clients, auditors and government still need to open exactly as they were made, and that's the part of the iceberg nobody has copied yet.
Generating quotes, reports, contracts and catalogue images from templates on your own infrastructure, with no per-seat cost and no document leaving the building, is the nearest-term win — and a cleaner POPIA story than sending client data to someone else's cloud service.
Deeper: Rented in dollars →Heavy users keep the flagship. Light users are where free alternatives — these and older ones like LibreOffice — save money first.
The macros, templates, plug-ins and file exchanges that must keep working. That list, not the feature checklist, decides whether you can switch.
Real documents, real deadlines, old licence kept as the safety net. Measure the files that broke, not the features that exist.
Where your team produces documents from data, ask whether an agent driving a free engine can do it without a seat licence.
Check the release history, the crash reports and whether the names changed. Momentum now says nothing about who's still maintaining it in April.
How PhotoCraft was built — the agent instructions, the automated rules, the honest scorecard and the risks — is in the tree as a full case study.
Deeper: PhotoCraft case study →Replacing expensive software one piece at a time, with the old system as a safety net the whole way.
Read briefing →Front office and back office: why the small visible part is never where the cost and risk sit.
Read briefing →Why your software vendors are set up to keep you renting — and what that costs in rand.
Read briefing →The first Craft app, read in technical detail: how it was built, what it measures, and what it's still missing.
Read the leaf →Agents made copying what software looks like cheap. They haven't made copying what it does under the waterline cheap — yet. Use the copies as leverage and as engines for agent-made documents now; move people onto them one team at a time, only when real files stop breaking.