I read a piece on Ars Technica this week that made me put my coffee down. A single developer, armed mostly with AI coding agents and a chip on his shoulder, is building open-source clones of Photoshop, Premiere and the rest of the Creative Cloud stack. Rough around the edges, by his own admission. But shipping, and shipping fast.
His line, quoted in the piece, is "software is over" — meaning the old model of charging rent on a GUI wrapped around some algorithms is finished, because the GUI and a fair chunk of the algorithms can now be generated faster than Adobe's product teams can hold a planning meeting about them.
I've been writing code since 1986. I've watched four or five "this changes everything" waves come and go. This one is different, and I think he's onto something, even if "over" is doing a lot of heavy lifting in that sentence.
Why this isn't just noise
Adobe's moat was never really the algorithms. Plenty of open-source image and video processing has existed for years — ffmpeg, GIMP, OpenCV, Blender. The moat was the enormous labour cost of building a polished, integrated, reliable product around all that: the UI, the edge-case handling, the file format support, the ecosystem of plugins, the muscle memory millions of users have built up.
AI coding agents are precisely the kind of thing that erodes labour-cost moats. Not because they're more creative than a human engineer, but because they're relentless and cheap. A solo developer with Claude or GPT-5-class tooling can now churn through the thousand tedious tasks — menu layouts, keyboard shortcuts, format converters, undo stacks — that used to require a team of twenty. The creative judgement is still human. The grunt work that used to take years now takes weekends.
That's the real story here, not whether this particular clone is good enough to replace Premiere today. It almost certainly isn't. It's clunky, missing features, and will be for a while. But the gap between "one person's weekend project" and "credible alternative" used to be measured in years of VC-funded headcount. Now it's measured in iterations of an agent loop.
The cautionary half of the tale
Here's where I'd push back a little on the triumphalism. Building a rough clone is the easy ten percent. The other ninety — reliability under real workloads, support, trust, integration with a creative professional's actual pipeline — is still brutally hard, and AI agents don't obviously compress that the same way. Adobe's biggest asset isn't code, it's that a studio can bet a production schedule on it not falling over.
So "software is over" is a punchy line for a blog post, less so as a strategy document. What's actually happening is that the cost floor for building a credible competitor has collapsed, not that the winner is predetermined. Incumbents with distribution, trust and workflow lock-in still have real advantages. They're just no longer protected by the sheer expense of building a competitor from scratch.
What this means if you build tools, not fine art
I build in a narrower lane — content and marketing automation rather than creative suites — but the dynamic is identical. RSSMasher, MarketMasher, Article2Video, the rest of the family: every one of them exists because there was once a real engineering cost to stitching APIs, content pipelines and publishing targets together in a way a non-technical marketer could actually use.
That cost is falling for me too, and for anyone else in this space. If I'm not actively using AI agents to compress my own build time, someone else is using them to come at my customers from underneath, the same way this Adobe-cloner is coming at Creative Suite. The defensible bit isn't the code anymore. It's the judgement about what to automate, the workflow design, the trust that the output is actually usable without babysitting it, and the relationships with the people using the tool.
My takeaway isn't panic, it's a reminder to keep checking which parts of what I sell are genuinely differentiated versus which parts are just "stuff that used to be expensive to build." The second category is shrinking fast, and pretending otherwise is how you end up being the incumbent in next year's Ars Technica headline.
Raw ambition, cheap compute, open-source code — that's the new base metal everyone's working with. Turning it into something people will actually pay for is still the only trick worth learning.
— Wayne