Will Schweitzer, CEO of Silverchair, moderated a candid conversation with Alison Mudditt (CEO of PLOS), Andy McGregor (VP of Product Management at Sage), Ann Michael (Independent Consultant), and David Crotty (Executive Director of Cold Spring Harbor Laboratory Press). Together, they explored how publishers can plan for a future in which more of their readers arrive through AI tools than through their own platforms.
Read the recap below or watch the recording here.
Setting the stage
Will opened with a story from Andy’s prep call. Sage had developed thirteen scenarios for the future of its platforms, ranging from new journal models to new publication processes, and when senior leaders ranked them by feasibility and desirability, the status quo came out on top. “It was a little irritating, I have to be honest,” Andy said, though he came to see the logic in it after colleagues across Sage reached the same result. “People are still going to need to come to our sites to access the articles,” he said. The difference now is that at least twelve other scenarios also need planning, and it isn’t yet clear which deserve the most investment.Ann wasn’t surprised, since subscription cycles and long publication timelines insulate scholarly publishing from rapid change. “Sometimes I feel like we’re an industry of frogs in water that’s getting hotter and hotter,” she said. “And I actually am very indebted to AI because I think some of the frogs are jumping out of the pot.” Alison offered a different starting point. As an open access publisher working under CC BY, PLOS has never been platform-centric, and she sees AI as accelerating a shift in how publishers serve their communities that was underway long before.
Designing for humans and machines
With non-human readers now outnumbering human ones on many platforms, Will asked whether publishers should keep investing in features for human readers. Ann cautioned against treating machine traffic as one category, since training bots, application-layer tools, and people working through AI assistants each have different needs. “I think it’s too early to say we don’t. And it’s probably too early to say we always will,” she said.David questioned how much of today’s platform serves readers at all, suggesting the journal homepage is “largely designed and maintained to make the owners of the journal and the editors happy more so than the readers.” Andy agreed that the focus is narrowing to the content page, where most readers arrive. “I think it’s going to be as much taking features away as it is concentrating on humans,” he said. Alison described PLOS’s move toward a distributed model, with articles on the PLOS platform and data, code, and other outputs in the appropriate repositories, which puts more weight on how metadata and trust signals travel between systems.
Rethinking how value is measured
As traffic declines and COUNTER metrics lag behind changing behavior, librarians are asking harder questions about subscription value. David framed the problem using foxes and hedgehogs. Most platform reading is fox reading, skimming abstracts and downloading PDFs that may never be opened, while hedgehog reading is the deep engagement of a researcher working through every figure in a paper that matters to their work. “AI tools can probably take care of a lot of the fox reading for us, but it can’t do the hedgehog reading,” he said, which leaves publishers needing new ways to show libraries the value of the reading that remains.Ann suggested thinking about it as a sales funnel. The fox readers at the top were unqualified leads, while the readers who arrive now come with a clear purpose, and the time saved for a library’s community is itself a form of value that current metrics don’t capture.
The panel also took up machine use of open content. Alison explained that PLOS is paid for providing a service to the research enterprise, not for recovering costs from downstream users, and that CC BY was designed for reuse. “Our authors are fine with reuse,” she said. “The problem is the lack of attribution.” She sees that gap as a design flaw in today’s AI tools rather than an inevitability, while David distinguished the legal requirement for attribution from the academic tradition of citation, which is the credit researchers care about most.
Experimenting without clear demand
Andy said Sage customers are asking about MCP endpoints, but few can say what they need, so Sage is experimenting broadly rather than rushing to build. “Unless we start to experiment, when it becomes clear what people want, we won’t be ready to move,” he said. David, working with limited resources at a research institution, has focused on what is unique. Cold Spring Harbor Laboratory Press now excludes its 50 years of laboratory manuals from licensing deals and is considering an MCP server built on that content that could be offered as part of a subscription.On near-term strategy, Ann recommended mapping possible scenarios, watching for the signals that show movement between them, and investing first in capabilities needed regardless of outcome. “If you really think that you’re going to have a plan that’s going to stay stable for three years, you’re not living in the world that I seem to be inhabiting right now,” she said. Alison added that insight alone isn’t enough. “You can have leading indicators, but the capability that goes with that is being willing to act on them,” she said, pointing to PLOS’s shift toward shorter, more iterative ways of working.
What survives
The discussion closed with the question of what role remains for publishers if the platform fades. Alison pointed to “trust, credibility, and judgment” as the functions to double down on. David argued that as AI makes it easy to produce accurate but incremental research, the journal’s role in designation and filtration becomes more valuable, giving readers a signal that experts have identified work worth their attention.Ann pushed the room to consider more disruptive possibilities, including AI research environments where provenance is built in from the start. “I think we really need to think about what does the nuclear option look like,” she said. The panel agreed that human judgment remains essential and that understanding how a result was reached matters as much as the result itself, or as David put it, “The output is irrelevant. The process is what matters.”
The session made clear that no one can yet predict where discovery is heading, but standing still isn’t an option. The organizations best prepared for what’s next will protect what makes them distinctive, experiment broadly enough to move when demand becomes clear, and build cultures that are willing to act on what they learn.
View all of the Platform Strategies 2026 session recordings here.