AI mastering has been "arriving" for close to a decade now — LANDR launched in 2014, and by 2016 was already being written up alongside CloudBounce as a challenger to human mastering engineers. Ten years on, the tools are faster, cheaper, and more widely used than ever. What they haven’t done, according to the most rigorous public test anyone has run on them, is actually catch up.
The test that mattered
In October 2024, musician and YouTuber Benn Jordan ran a double-blind listening study on his own 2017 track "Starlight," sending it to a mix of AI mastering plugins, AI-powered online services, and professional human mastering engineers, then asking 472 people to rate the randomized, unlabeled results. Twelve semi-finalists were narrowed to seven for the final round. LANDR — one of the most widely used AI mastering services on the market — was disqualified before that final round because it simply didn’t produce a master good enough to compete. Of the seven finalists that did make it through, the top two spots went to human engineers: Max Hosinger placed first, Ed the Soundman second. The best-placed AI or algorithmic tools — Matchering 2.0, an open-source option, and Ozone paired with Neutron — landed third and fourth.
What AI mastering is actually good at
None of that means AI mastering is useless — it means the "AI already replaced mastering engineers" framing gets the current state of things backwards. Where these tools do genuinely help is speed and accessibility: LANDR’s entire pitch since 2014 has been giving independent artists a fast, cheap, competent master without booking studio time, and iZotope’s Ozone has become a real part of many producers’ own mixing and mastering workflow, not as a replacement for a human ear but as a tool a human is actively steering. That’s a meaningfully different claim than "AI now rewrites the rules of mastering" — it’s AI becoming a genuinely useful tier below professional mastering, not a replacement for the top of it.
The company building it has said this from the start
This isn’t a new admission forced by an unfavorable study, either. Back in 2016, LANDR co-founder Justin Evans told Sound on Sound directly: "there are many things a mastering engineer will do that we will never be able to do." Eight years before Jordan’s blind test confirmed it empirically, the people building the technology were already saying the same thing about its ceiling.
So what’s actually changed
What’s real about the "AI is rewriting mastering" story isn’t that the results caught up to a professional engineer — Jordan’s study suggests they still haven’t. It’s that AI mastering’s floor has risen enough to be a legitimate, fast, low-cost option for the enormous number of tracks that were never going to get professionally mastered at all: demos, quick releases, bedroom-producer catalogs, anything where the alternative wasn’t a human engineer but no mastering at all. That’s a real shift in how much music gets a competent master. It’s a smaller and more honest claim than "AI is rewriting the rules" — and, ten years into this technology existing, the people who make it have been saying so the whole time.