Splice CEO on Where AI Stops in Music Creation
Splice CEO Kakul Srivastava says AI will never replace human creativity, and she wants Splice to build around that boundary.
She set out that position in interviews with The Verge, including a Decoder conversation with Nilay Patel published March 24, 2025 on where to draw hard lines around AI in music, and a follow-up interview published Oct. 3, 2026. The Verge
Srivastava joined Splice as CEO from Adobe after serving on its board since 2021. Founder Steve Martocci moved at that time to Executive Chairman and Chief Strategy Officer. She had held the CEO job for a little over six months as of December 2022. Her earlier roles included Flickr, Yahoo, GitHub and Adobe. The Verge She earned a BSc in Mechanical Engineering from MIT and an MBA from UC Berkeley's Haas School of Business. She was named to TIME's 100 Most Influential People in AI in 2025.
Splice is known for its sample library and creator tools. Its Sounds Plugin, now in beta, puts millions of royalty-free, high-quality samples, pre-recorded sounds cleared for reuse, inside a DAW, the editing software where music is arranged and mixed. Users can search, preview, drag and drop sounds, and make variations without leaving a project. The company acquired Spitfire Audio. Splice samples were used in Lisa's "Money" and Sabrina Carpenter's "Espresso."
Splice operates remote-first. Its applied research and education work includes Vice President of Applied Research Alejandro Koretzky and Director of Business Development, Education Meredith Allen, who both represented Splice at the 2026 NAMM Show.
Srivastava names process, not software, as her most indispensable tool. She called it "a well-written document and a real conversation about it." She said she is cautious about AI-written documents because "once you can't tell whether there is a person behind the words, the conversation becomes different and worse."
On lighter questions, she named Instagram as her happy place online, Antakshari as the game she remembers most fondly, and a cup warmer as her most disappointing gadget.
The broader context here is worth spelling out. In my view, Srivastava is pointing to two control points. One is at the level of the finished sound, which inputs and outputs are allowed. The other is at the level of coordination, which plans still show accountable human intent. That matters because in-DAW search, preview and variation tools shorten iteration time. Speed helps. Clear attribution holds the system together.
In my view for builders, the limit is not whether models can make plausible music. They can. It is whether musicians, producers, educators and rights holders accept the workflow. A remote-first sample platform depends on trust in its library, clear licensing and low latency in a session. Keeping a human author visible in planning and human performance audible in tracks preserves that trust. That leaves room for better retrieval, better variation tools and better teaching, without handing authorship to autocomplete.
