Most artists working with AI ask whether a machine can make art. Jonas Lund has spent more than ten years on a sharper question: if you could measure what makes art succeed, could you manufacture it? And what would that do to the people involved?
His newest work, Performance Review (2026), puts that question in an office. An AI agent acts as manager, directing human workers to produce artworks and then write the context that presents them as art. It opened as a solo show at OFFICE IMPART in Berlin (April 16 – May 29, 2026), developed with ZKM Karlsruhe. According to a new interview with Lund by Anika Meier for Right Click Save (published Sept 15), an expanded version runs for about six months at the Busan Museum of Art as part of its reopening exhibition on the future of museums.
The long experiment
Performance Review makes more sense as the latest chapter of a very consistent practice:
- The Fear of Missing Out (2013) mined art-world data to generate instructions for artworks that were statistically likely to succeed, then carried them out.
- Flip City (2014) produced digital paintings on canvas, in dialogue with the market logic of flipping.
- Strings Attached (2015) came with contracts restricting what buyers could do with the work.
- Studio Practice (2016) gave assistants a 300-page manual for making paintings, with an advisory board scoring the results.
- Jonas Lund Token (JLT) (2018–ongoing) gives token holders a vote on the artist’s decisions. Jonas Lund Futures sells forward contracts that convert into paintings at a later date.
Each piece takes something the art world does quietly (predicting taste, outsourcing labor, turning careers into assets) and makes it a visible, legally binding system. Performance Review adds the step that has become normal elsewhere: handing management to software.
Why the AI-manager version hits harder
Earlier works were satirical projections. The algorithm and the manual were Lund’s own inventions. An AI agent assigning tasks and judging human output is no longer a thought experiment. It’s how a growing number of workplaces run. Framing it as a “performance review” makes clear that the subject isn’t just art but everyone’s work under automated oversight.
It also inverts the usual worry about AI art. Here the machine doesn’t make the work. Humans make it, and the machine manages them. Whatever gets exhibited comes from people working to someone else’s metrics, which is arguably a fair description of much art made for the market.
Where it ends up
Lund’s conclusion in the interview is the most interesting part, because it goes against the grain of his own methods. Art, he says, “can be optimized for visibility, markets can manufacture scarcity,” but none of those systems can finally explain why one work matters. After ten-plus years building systems to capture artistic value, his answer is that “it’s impossible to quantify,” and that this is the magic of art.
That doesn’t undermine the practice. It’s what the practice was always testing for. The optimization systems are the method, and their failure to explain why a work matters is the finding.