Who Is Kakul Srivastava and Why Her Voice Matters in AI Discourse
When Kakul Srivastava talks about how artificial intelligence is reshaping communication, she speaks from two vantage points that rarely intersect. She runs Splice, the sample platform whose loops and one-shots have been woven into tracks like Lisa's "Money" and Sabrina Carpenter's "Espresso." Before that, she spent years in executive roles at technology companies, watching product teams wrestle with the same automation questions now confronting every knowledge worker on the planet.
That dual perspective matters. The debate over AI-generated email tends to be dominated by productivity consultants and enterprise software vendors, parties with an obvious incentive to frame automated correspondence as progress. Srivastava arrives from the creative economy, an industry where the difference between a genuine exchange and a synthetic one is not an abstraction. Musicians know what happens when a human touch gets flattened into something efficient but hollow.
Her argument, aired in a conversation with The Verge, is blunt: AI emails are killing conversations. Not merely cluttering inboxes. Killing the thing that makes correspondence worth having in the first place.
The Core Argument: AI Emails Are Replacing Real Conversations
Srivastava's claim lands against a backdrop of near-universal adoption. Microsoft has embedded Copilot into Outlook and Teams; Google has folded Gemini into Gmail and Workspace; a sprawling ecosystem of startups offers to draft, summarize, and auto-reply on a user's behalf. Email, the oldest surviving medium of professional life, has become the primary testing ground for generative AI in the office.
Read next Laika's Wildwood: Stop-Motion Fantasy at TIFF 2026The empirical picture is messier than the marketing suggests. McKinsey's research on generative AI in the workplace has consistently found that the technology delivers its clearest productivity gains in tasks involving summarization, drafting, and information retrieval — precisely the tasks that constitute most email. But the same research repeatedly flags a trade-off: speed rises while quality, judgment, and trust can degrade when outputs are accepted without scrutiny. A 2024 MIT Media Lab study on AI-assisted writing found that participants who used large language models produced more uniform text and reported weaker feelings of ownership over their work. The writing got faster. It also got more generic.
That is the crux of Srivastava's concern. When both sides of an exchange delegate composition to a model, the correspondence becomes a handshake between two autocomplete engines. Replies arrive instantly and say nothing. The friction that once forced people to clarify, negotiate, and commit disappears — and with it, the shared understanding that meetings, deals, and creative partnerships depend on.
This is not a fringe worry. Organizational communication researchers have spent years documenting how mediated messaging already erodes nuance. Add a generative layer that optimizes for plausible-sounding filler, and the erosion accelerates. The inbox fills up. The conversation empties out.
How Splice Became a Cornerstone of Hit Music Production
To understand why anyone should care what a sample-platform CEO thinks about email, consider what Splice actually is. The company operates one of the largest catalogs of royalty-free samples and loops in the music industry, a library that independent producers, bedroom beatmakers, and Grammy-winning songwriters draw from daily. Its sounds have surfaced in chart-dominating records, including work by Lisa and Sabrina Carpenter — placements that put Splice's fingerprints on some of the most-streamed pop music of the past several years.
That footprint gives Srivastava a specific kind of authority. Splice sits at the intersection of technology and creative labor. Its business depends on producers finding raw material that feels human enough to build on — a bass line, a vocal chop, a drum pattern with texture and imperfection. The entire value proposition rests on the premise that human-made fragments, recombined by human hands, produce something a machine cannot.
Srivastava has therefore spent her tenure navigating the AI question from the inside. Splice, like every music technology company, faces pressure to automate discovery, generation, and recommendation. The temptation to let models do more of the creative work is constant. Her position — that efficiency has limits, and that some exchanges must remain irreducibly human — carries weight precisely because she is not a Luddite. She runs a technology company. She has shipped AI features. She is arguing for boundaries from within the house, not from outside it.
The Broader Tension Between AI Efficiency and Human Creativity
Srivastava's email critique is one expression of a larger argument playing out across the creative industries. The music business has become a live experiment in what happens when generative systems enter a domain long assumed to be immune to automation. Streaming platforms now host AI-generated tracks by the tens of thousands. Production tools offer stem separation, mastering, and mixing assistance that once required a trained engineer. Each advance eliminates a task; each elimination raises the same question. Which tasks were merely labor, and which were the work itself?
A responsible AI adoption framework starts by sorting tasks into categories: automatable, assistive, and protected. Email triage, calendar scheduling, and meeting summarization belong in the first bucket. First-draft composition and data synthesis fit the second. Negotiation, creative direction, and relationship-building belong in the third — and should stay there. Organizations that blur these lines in pursuit of headcount savings tend to discover the cost later, in the form of disengaged teams and decisions made without real deliberation.
Music technologists who have weighed in on AI-mediated creative work tend to echo this triage logic. The consensus is not that AI should be barred from creative tools — it is already embedded, and usefully so. The consensus is that automation should absorb the mechanical and leave the expressive untouched. When a model drafts your email, it is not saving you time on a mechanical task. It is making an expressive choice on your behalf, and doing it badly.
What Music Tech Leaders Can Teach Us About Responsible AI Adoption
The music industry's experience offers three lessons that translate directly to how enterprises should handle AI in communication.
First, provenance matters. Splice built its reputation on transparent, licensed samples with clear origins. Producers know what they are working with. The same standard should apply to AI-generated correspondence: recipients deserve to know when a message was composed by a model rather than a person. Several jurisdictions are already moving toward labeling requirements for synthetic content. Email is next.
Second, the human layer is the product. The reason a Splice loop ends up in a hit record is not that it is convenient. It is that a human made a creative decision that another human found useful. Strip the human contribution from an exchange entirely, and you strip the value with it. Acknowledge receipts within a day. Delegate scheduling to a bot. Do not delegate the thinking.
Third, adoption should be judged by outcomes, not output. The MIT findings on homogenized AI writing and McKinsey's warnings about unverified AI outputs point the same direction. Measure whether decisions improved and relationships strengthened — not whether reply volume increased.
Conclusion: Rethinking How We Let AI Into Our Workflows
Kakul Srivastava's critique of AI email is not nostalgia. It is a warning from someone who has built a career on the productive marriage of technology and human creativity — and who knows where that marriage breaks down. The tools are here, they are useful, and they are not going away. The question is where the line sits. Draw it badly, and the inbox becomes a hall of mirrors, everyone corresponding with everyone else's chatbot. Draw it well, and AI handles the noise while people handle the meaning. That is the standard Splice applies to samples. It is the standard the rest of us should apply to our conversations.
Related coverage
Source: The Verge
