Last week, a developer named Brandon Thomas released seven free, open source apps under the ArtCraft name. They are clean room rebuilds of Photoshop, Illustrator, Premiere, Lightroom, After Effects, InDesign and Acrobat Pro, written in Rust, largely with Anthropic's Claude Opus 5.5. They run natively on Windows, macOS and Linux, several also run in a browser, and the code is MIT and Apache 2.0 licensed, so anyone can take it and build on it.
His motive was ordinary. He was fed up with Adobe's subscriptions, including a fee of 50% of the remaining contract if you cancel an annual plan after the first 14 days.
Most coverage asked whether Adobe should be worried. That is the less interesting question. The better one, especially for a continent whose developer base Boston Consulting Group puts at about 4.7 million and growing about 21% a year, is this, what does it mean for how developers learn, and for how software businesses start?
What: Seven clean room rebuilds of Adobe apps, from Photoshop to Acrobat Pro
Built with: Rust, largely written by Claude Opus 5.5; browser builds via WebAssembly
Licence: MIT and Apache 2.0, free to fork and build on
Status: A “very rough alpha”, estimated at 25% to 45% of Adobe’s features
Why: Its maker was fed up with Adobe’s subscriptions and cancellation fee
Seven careers, one person
Look at what ArtCraft covers. Raster image processing, vector geometry, video timelines, colour science, motion compositing, page layout and the famously awkward PDF format. In the traditional software industry, each of those is a specialism people build whole careers in. Adobe built its suite over decades, with large specialist teams for each product.
One person now has working versions of all seven, even though they are not yet complete. In the past, it would have taken years to learn enough about each field to build something useful. The time and effort have now been greatly reduced.
That tax always hit African developers hardest. Most local job markets have few openings for graphics engineers, compiler specialists or video-codec experts, so few people ever got to learn on the job in those fields. Paid courses are priced in dollars. Senior mentors are scarce and stretched. The usual path ran through outsourcing work on someone else's product, in someone else's specialism. When the learning tax falls, a developer in Kumasi or Kigali can explore a field before any employer offers a job in it.
The curve didn't flatten. It moved.
The easy version of this story is "anyone can now build anything". ArtCraft itself disproves it.
Thomas calls the release a very rough alpha. Estimates of how close it comes to Adobe's features range from about 25% to 45%, and early users report problems with basic editing, keyboard shortcuts and performance. AI made the first half of the job almost free. The second half is file compatibility, edge cases, speed on real hardware, and the hundred small behaviours professionals depend on. That is where the project is stuck.
So the learning curve hasn't gone. It has shifted from writing code to judging it:
- Specifying. The bottleneck is describing exactly what the software must do, including the cases nobody mentions.
- Reading. Spotting what is wrong in generated code is now worth more than typing it.
- Testing. Proving the software works, on a cheap phone and a weak connection, is the job.
- Knowing the domain. No model knows your users, your tax office or your market better than you do.
The old curve was steep at the start, months of syntax before you could build anything. The new one is gentle at the start and steep at the end. That reverses where beginners get stuck, and most training programmes haven't caught up.
The old curve was steep at the start. The new one is gentle at the start and steep at the end.
The hard language is no longer the hard part
ArtCraft's choice of Rust is easy to miss. Rust is known for being fast, memory safe and hard to learn. Its compiler rejects code that other languages accept, so teams have usually adopted it only once they had senior engineers who could get through that learning curve.
If AI can carry a developer through that strictness, the best engineering tools are no longer reserved for well staffed teams. That matters in Africa, where software often has to run well on older laptops and low cost Android phones. ArtCraft's browser builds use WebAssembly, which lets code compiled from Rust run at near native speed in a browser. That is the kind of performance engineering that used to need years of experience before anyone attempted it.
Build to learn, not learn to build
The most practical change is the order. The old route was to learn the fundamentals, then build. The new route runs the other way, build something real, then study why it works, where it breaks and what the AI got wrong.
That turns the clean room rebuild into a curriculum. Rebuilding a known tool, without copying its code or assets, as ArtCraft says it did, gives a learner a finished target to measure against, plus a portfolio piece employers understand. The legal line matters. Recreating features and behaviour is one thing. Copying code, icons or trademarks is another.
But a rebuild is only a training ground. The real test is building what doesn't exist yet, and that is where the second curve comes in.
The business curve bends too
For most African founders, the expensive part of starting a software company was the first version. A team, months of runway and often a raise before a single customer could try the product. Since the global funding slowdown of 2022, that raise has been harder to find on this continent than almost anywhere.
ArtCraft points to a different sequence. When one person can build a credible first version in weeks, the first cheque is no longer for building. It is for finding out. A founder can put a working product in front of fifty real customers before talking to an investor, and arrive at that meeting with usage data instead of slides.
Three things follow, and they are less comfortable than they sound.
Building stops being a barrier. If you can rebuild Adobe's interface in weeks, so can the next founder, and so can a well funded foreign competitor. What can't be copied quickly is what sits around the code, distribution, a trusted brand, payment rails that actually work, regulatory approvals, and a support line in the customer's language. African startups have always had to win on those. Now they are almost the only things that count.
Pricing gets more local. ArtCraft gives its software away and sells optional AI credits. When building costs fall, small ticket pricing becomes viable, weekly plans, local currency, mobile-money billing. That suits African markets far better than a $59.99 monthly dollar subscription.
Learning speed becomes the business. The team that learns fastest from customers wins, because the team that builds fastest no longer has an edge. The learning curve and the business curve turn out to be the same curve.
Here are three apps where Africa's constraints make both worth climbing.
The first cheque is no longer for building. It is for finding out.
Three apps worth climbing the curve with
A design suite paid in naira
Adobe's All Apps plan costs about $59.99 a month on an annual plan, according to PetaPixel. When foreign exchange ran short in 2022, UBA, Zenith and other Nigerian banks cut naira-card international spending to $20 a month, and some suspended it completely. A creative tool priced in local currency, paid by mobile money or transfer, and fully usable offline would serve designers from Lagos to Lusaka. ArtCraft's open code is a starting point.
Invoicing that speaks fluent FNE
Côte d'Ivoire is moving businesses onto the Facture Normalisée Électronique, a standardised electronic invoice issued through the tax authority's platform or connected to it by API. Under the 2025 rollout, paper invoices stopped qualifying for VAT deductions from December 2025. Deadlines have since been adjusted, so check the DGI's latest notice. Vendors such as Sage already sell FNE features to bigger firms. The opening is a French first tool for the hairdressers and hardware stores of Abidjan and Bouaké.
An office suite in Amharic
Global office suites ship English first and treat many African languages as an afterthought. For a civil servant in Addis Ababa working in Ge'ez script, that is daily friction. African language AI is no longer starting from scratch. The Masakhane research community, founded at the Deep Learning Indaba in 2019, publishes open datasets and models, including an entity recognition model on Hugging Face that covers 21 African languages, among them Amharic, Hausa and Wolof.
The trap, understanding debt
There is a cost to all this, and African tech should name it early.
When code arrives faster than understanding, developers build up understanding debt, software in production that nobody on the team can fully explain. That is tolerable in a photo editor. It is dangerous in an invoicing system the tax authority audits, or a payments flow the central bank supervises.
There is a quieter risk too. The routine tasks juniors used to learn on, such as small bug fixes, simple pages and boilerplate, are the first ones AI absorbs. If those rungs disappear, the industry has to build new ones on purpose, code review as teaching, testing as a first job, and rebuild projects that someone senior actually checks.
ArtCraft shows that one person can now get 40% of the way to a world class product in weeks. The developers who win the next decade will be the ones who learn how to cover the other 60%, and Africa's constraints make that the best place on earth to practice.
