A new AI-powered word processor enters the market with a clear message: to help writers, not replace them. Marker, a London-based startup founded by Jon Steinback (former DeepMind creative lead) and Ryan Bowman, has just closed a $13 million seed round led by Index Ventures, with participation from LocalGlobe and angel investors such as Steve Newman (co-founder of Writely, later Google Docs), Slack CTO Cal Henderson, and Hugging Face’s Thomas Wolf.

The tool bills itself as a “reimagined word processor,” built to support the creative process: from ideation and rough drafts to revision and collaboration with co-writers and commenters. Its stated goal is to offer an alternative to “AI slop,” the wave of text generated by Large Language Models that, according to Synthesia CEO Victor Riparbelli, is eroding the quality of automated writing. «We’re in a moment where people get to choose the future of writing, and I believe they will choose something that values the craft, rather than the slop brutally eroding it,» Steinback said.

In some respects, the formula echoes the transformation that Figma brought to collaborative design or Notion to idea organization. Georgia Stevenson, partner at Index Ventures, stated: «Creative people deserve tools that understand their craft. Writing—the most universal creative act—has been left behind, stuck between legacy word processors and automation tools.» Marker proposes an approach where AI does not write on behalf of the user but acts as an extension of their intent, keeping them “in the flow.”

Beneath the anti-slop rhetoric, however, lies a question that is critical for AI‑RADAR readers: where does the data reside, and who controls it? Marker has announced no on-premise or self-hosted deployment option. Like the vast majority of AI-enhanced productivity tools, it presents itself with a cloud-centric architecture—a factor that, regardless of user experience quality, represents an insurmountable barrier for sectors where data sovereignty is non-negotiable: law firms, financial institutions, healthcare, public administration, and any organization bound by GDPR or equivalent regulations.

This gap between functional innovation and infrastructural control creates an ever-deepening divide in the AI writing tool market. On one side, there is a race to perfect the interface and the capacity to keep the writer “in the zone”; on the other, an unmet demand for solutions that can operate air-gapped or on proprietary infrastructure, without prompts and text transiting through external servers. It is a trade-off that startups focused on user experience tend to overlook in their early stages but that becomes the true adoption criterion for enterprise decision makers.

The very existence of Marker, with its focus on artisanal text quality, could paradoxically accelerate the demand for on-premise alternatives. If an organization recognizes the value of an AI writing assistant that does not generate slop, it will be all the more driven to seek a solution that combines those capabilities with the security of local deployment. History shows that tools born as consumer products (think of office suites) only reached the enterprise world after integrating granular data management options and hybrid installation models. Marker, at least for now, has not charted that course.

For those evaluating the adoption of LLMs in professional writing, this case is emblematic: fighting slop is necessary but not sufficient. Without a clear answer on deployment, any AI word processor risks remaining confined to the individual creative or the team that can afford to accept cloud mediation. And this is where the debate shifts from tokens to hardware control—a domain that AI‑RADAR will continue to cover, analyzing the infrastructures and stacks that make truly sovereign AI writing possible.