2026 is already among the highest-volume years ever for audio plugin and app releases, with nearly three months still to run. The global audio plugin market was $1.26 billion in 2025 and is projected to reach $2.51bn by 2033.
As Synth Anatomy’s editor puts it: “It’s not because everyone suddenly took a crash course in coding.”
He has published a long piece on what AI-assisted plugin development is doing to the market and — unusually — to his own job. It is worth reading in full, and worth engaging with here because the same pressure applies to this site.
What it used to take
The historical barrier is the part people forget. Waves shipped the first audio plugin in 1992 for Digidesign’s Sound Tools/Pro Tools. Steinberg’s VST followed in 1996 and became the standard; AU, AAX, CLAP, LV2 and AUv3 came later.
To turn an idea into a working plugin you needed audio expertise, programming (C++ and friends), digital signal processing, mathematics, and knowledge of an audio framework. Four or five deep specialisms, in one person or one small team. That stack is why the plugin market was small, slow, and populated by companies that lasted decades.
It is also why a plugin used to carry information. A product existing meant somebody had cleared that bar.
What the flood actually breaks
The editor’s own framing covers the good and the bad, and the section headings tell you where the argument goes: the open-source problem, security & future-proof, the consequences for the others: visibility, more of the same, the consequences for the others: prices, and my coverage.
Two of those are the ones that matter to anyone making or buying this software.
Visibility. When ten plugins ship a week, the signal that a product exists becomes worthless, and attention goes to whoever markets hardest rather than whoever built best. The developer who spent two years on a filter is now in the same list as something generated over a weekend, and the list is sorted by recency. That is a direct transfer of value from people who did the hard version to people who did not.
“More of the same.” This is the subtler cost and the one we have measured elsewhere. A generated plugin is, by construction, assembled from patterns in what already exists — so the flood is not a flood of novelty, it is a flood of competent variations on solved problems. Compressors, saturators, reverbs. Design Creativity Bench found exactly this shape in LLM-generated UI: appropriateness near-perfect, originality and range far below the human baseline. There is no reason audio would differ.
The licensing argument is the sharpest part
The open-source problem is the heading, and it is the same objection Synth Anatomy raised about paid AI-generated firmware for the MicroFreak:
“I don’t like paying for these hacks, since the current wave of custom firmware is AI-based and relies on models trained on open-source projects. That way, they should be free and open-source.”
And it is the same thing that surfaced today when Émilie Gillet open-sourced Mutable Instruments Beads under MIT, with the observation that the first beneficiaries were scrapers that will not honour the attribution the licence requires.
Three separate stories in a week, all circling one unresolved question: the audio open-source corpus — JUCE-adjacent code, Mutable’s firmware, decades of published DSP — is the training material, and permissive licences were written for human reuse with credit, not for statistical absorption without it. Nobody has a mechanism for that, and the people most affected are the ones who published generously.
What a flood does to any filter
The editor is explicit that the volume is changing what he can cover, and the structural point generalises past one outlet. Anyone trying to report every release drowns; anyone who filters has to defend the filter.
What changes in a flood is the nature of the work. Selection stops being about finding the interesting thing and starts being about excluding the plausible one — and plausible is precisely what a generated product is best at. A thing that is competent, complete, documented and derivative is far harder to rule out than a thing that is bad.
The consequence runs all the way to the buyer: existence stops being evidence. A plugin being released, having a website, and having a feature list no longer indicates that anyone solved anything. The things that still carry information are a named author with a history, a published technique, an open repository, measured figures, or someone you trust saying it sounds good.