By This Hour Technology Desk

Kakul Srivastava, the chief executive of music-creation platform Splice, has set out a sharply skeptical view of AI-generated workplace writing: it may not simply save time on emails and documents, but alter the conversations that follow them. Her concern is not limited to whether a generated draft is polished or useful. It is about whether colleagues believe there is a person behind the words, and what happens to deliberation when that belief becomes uncertain.

In comments published by The Verge, Srivastava described a written proposal followed by a real discussion as a crucial working practice at Splice, which she characterized as remote-first. Without the chance encounters of a shared office, she said, the company depends on someone working through an idea in writing and other people engaging with it directly. Her argument is that AI authorship can disrupt that chain: if the apparent author is unclear, the response becomes a different, and in her view poorer, kind of exchange.

The remarks place a workplace question alongside the wider argument over generative AI. Much of the attention around AI writing concerns speed, drafting help and the quality of the finished text. Srivastava’s concern instead centers on the social meaning of a document. A memo, email or proposal is not only a container for information in her account; it also represents somebody’s judgment, effort and willingness to stand behind a position. If that connection is obscured, she suggests, the discussion may lose something important even where the prose itself appears competent.

A document is only the beginning of the exchange

Srivastava’s formulation gives particular weight to the sequence of writing and response. The document matters because it records that somebody has considered a question enough to present a view. The conversation matters because colleagues can then challenge the reasoning, test assumptions and work through disagreement. In her description of Splice’s internal process, productive outcomes have often begun with precisely that pattern: a considered document and a group prepared to argue with it.

That account also explains why she treats uncertainty over authorship as more than a disclosure problem. If readers are unsure whether the sender composed the text or merely supplied a prompt, they may assess the message differently. They may be responding to a person’s considered position, to generated wording that has been lightly reviewed, or to some mixture of the two. Srivastava’s point is that the uncertainty itself changes the terms of the interaction. The recipient is no longer simply engaging with the argument in front of them, but judging what degree of ownership and thought lies behind it.

Her warning is deliberately broad enough to cover more than email. She referred to AI-written documents, a category that can include the written materials used to initiate workplace discussion. Yet the emphasis on conversation keeps the criticism from being a simple objection to automated text. She did not say in the reported comments that every use of AI assistance makes a document worthless. Rather, she expressed caution about writing generated by AI when it becomes hard to determine whether a human being is behind the words.

That distinction matters. Srivastava is describing an organizational risk, not offering a reported measurement of how often AI-generated correspondence harms decisions or relationships. The available account does not set out a company policy on disclosure, identify particular messages that caused problems, or establish how widely employees use such tools. It presents the chief executive’s philosophy of communication and her concern about its consequences.

Remote work gives the concern added weight

Splice’s remote-first model is central to Srivastava’s reasoning. She contrasted it with an office environment where people can meet informally in hallways. In a distributed company, a written document can carry more of the burden of beginning a substantive discussion. It can offer a common starting point for people who are not physically together, while the follow-up conversation gives them a chance to question, refine or reject the proposal.

For Srivastava, that makes authorship part of the infrastructure of remote work. The issue is not nostalgia for a particular style of memo. It is the need for colleagues to know what they are being asked to react to and who is accountable for the thinking. Her comments imply that a remote team may have fewer informal cues available to establish that connection. In that setting, a document whose human origin is uncertain can create distance rather than bridge it.

There is an inherent tension in that view. AI tools can be used to turn fragments of an idea into a structured draft, which could make it easier for someone to put an issue before colleagues. But Srivastava’s argument is that a faster first draft is not necessarily equivalent to the intellectual work a group needs from its members. The value she assigns to writing lies partly in the act of thinking before the discussion begins, not just in producing a readable artifact for others to consume.

Her position also puts the burden on leaders and teams to decide what they expect from internal writing. The reported interview does not prescribe a rule, such as banning assistance or requiring labels on every AI-touched message. Nor does it offer a test for separating extensive generation from ordinary editing. Those unanswered practical questions are significant because the boundary can be difficult to define. Srivastava’s comments identify the trust concern but do not supply a detailed operating framework for resolving it.

Her reservations extend from workplace text to music tools

The view is consistent with the approach to AI in music that Srivastava has publicly associated with Splice, as described in the interview. Splice provides samples and creative tools for music makers, placing the company near a contentious debate about generative systems and creative control. The source account says Srivastava has sought to steer the company toward AI products that keep people involved in the creative process rather than reducing music creation to a fully automated output.

She explicitly linked her concern about written communication to music. The comparison does not mean that workplace documents and songs raise identical questions. Instead, it suggests a common principle: tools can assist human activity, but a system may become less valuable to her when it removes or conceals the person making choices. In both settings, Srivastava appears focused on the relationship between an output and the human judgment that shaped it.

That perspective is informed by a career that has crossed technology and creative work. The Verge described Srivastava as having held executive roles at Flickr, Yahoo, GitHub and Adobe, and as having been part of the early team at Tiny Speck, the company that later became Slack. The interview also recounts her reflection on Tiny Speck’s game, Glitch, which she described as creatively ambitious but unable to account quickly enough for major changes in technology and user habits.

That history does not prove her current assessment of AI-written communications. It does, however, provide context for why she frames technology choices around what they preserve or displace. In the same interview, she described online video as a way to reconstruct recipes associated with her late mother, illustrating a more affirmative account of technology: a tool can carry knowledge between strangers and recover a personal connection. Her view is therefore not a blanket rejection of digital tools, but a selective judgment about what they should enable.

The unanswered question is how teams preserve trust

Srivastava’s remarks leave open several questions that organizations adopting AI writing tools will have to answer for themselves. How much human review makes a draft genuinely owned by its sender? When does assistance become authorship? Does a reader need to know how the text was produced, or is it enough that the named sender accepts responsibility for it? The interview does not resolve those questions, and it supplies no evidence that one answer will suit every workplace.

It also does not establish a causal link between AI-written email and poorer conversations across companies. Srivastava offered a judgment from her experience and values as a technology executive leading a remote-first business. Her concern may be especially relevant where documents serve as the basis for consequential debate, but the source material does not compare different kinds of workplace communication or report the perspectives of Splice employees, customers or other executives.

Still, the argument is clear. Srivastava sees the useful unit of work not as a generated message sent with minimum effort, but as a human being thinking through an issue, putting forward a position and remaining present for the disagreement that follows. In that model, a tool that makes language easier to produce can become counterproductive if it makes the underlying commitment harder for others to see.

The report is based on a single published interview and has not been independently corroborated. The available material supports Srivastava’s stated view and the biographical and company context reported there, but it does not independently verify the broader effects of AI-written documents on workplace discussion or establish how Splice applies her approach in day-to-day practice.

For further context on this subject, see At Least 14 Reported Dead as Floods and Landslides Hit Nepal.

Reporting notes

What is confirmed: The available account is a published interview describing Srivastava’s views on AI, remote work and creative tools.

Why this matters: Her comments frame AI writing as a question of trust, ownership and deliberation, not only efficiency.

What remains unclear: There is no independent evidence in the supplied material measuring the effect of AI-written documents on workplace conversations or detailing Splice policy. This report is based on one source and has not been independently corroborated.

Sources