How much time have we lost lately preparing manuscripts for journal submission? Reformatting references for each new venue, filling in metadata fields the system could extract itself, drafting cover letters, ticking compliance boxes, uploading files in the order the interface demands - and then repeating the whole performance at the next journal when the first one says "no" after fourteen months of deliberation? What is the most long-suffering paper on your CV, the one whose references were already ageing gracefully by the time it finally appeared in print?

Submission volumes keep growing while the pool of available reviewers and editors has not kept pace, and the system has responded to this pressure the way overstretched bureaucracies always do: by adding more forms, more declarations, more fields that ask for the same information in slightly different ways across three consecutive screens. Many colleagues have moved to arXiv or SSRN to sidestep the wait and get their work visible immediately.Their logic is understandable, yet what these platforms offer is visibility before any institutional evaluation, and visibility alone is not what makes research findings citable as evidence, actionable in policy, or trusted as a basis for further work. The queue gets longer, the workarounds multiply, and the architecture underneath it all remains exactly as it was designed in 2001, apparently awaiting the return of its original authors before anyone can touch it.

A genuinely interesting way out of this dead-end was proposed recently by Luciano Floridi, philosopher of information and digital ethicist at Yale and the University of Bologna, in his article "The editor's signature: a proposal for AI-born journals," now available on SSRN. Floridi introduces two concepts that reframe the whole situation. The first is the AI-assisted journal: a venue that adds AI tools to an existing workflow while leaving every incentive intact to keep generating new procedural requirements until the saved hours are quietly consumed by the next compliance update. The second is the AI-born journal: a venue whose entire architecture was designed from scratch around what AI agents can reliably do today. Floridi's central interest is in specifying that architecture in enough concrete detail to show how AI can absorb everything surrounding the editorial act, so that editors are left doing what they are actually there to do: weighing reviewer opinion against the state of the field, deciding which objections are decisive, and accepting personal accountability for the conclusion.

The architecture works as follows. The author submits the manuscript once, in whatever format they wrote it, with no reformatting for the journal's house style. An agent extracts metadata directly from the file and the author reviews and corrects the extraction in about ten minutes, which is roughly the time currently spent on one metadata field that requires entering the same information twice for reasons nobody alive can reconstruct. A second agent runs integrity checks and produces a report for the editor. A third agent generates a ranked list of candidate reviewers drawn from co-citation patterns, ORCID profiles, OpenAlex records, and recent reviewing load; the editor selects from it, and the agent manages invitations and follow-ups. Reviewers receive a structural pre-analysis that maps the paper's claims, summarises its methodology, and situates it against the cited literature, so each reviewer can move directly to the judgment only they can make. When the reviews arrive, a fifth agent synthesises them into a draft decision letter; the editor reads the original reviews alongside the draft, revises what needs revising, and signs.

The governance layer presumes that every agent output is reversible by the editor without friction, and every override is logged for the editorial board to inspect on a rolling basis, because "the editor remainsin charge" is only a convincing statement when the architecture is willing to show exactly how.

The philosophical foundation underneath this design explains which functions the architecture deliberately preserves. Floridi's core claim is that a journal's irreducible contribution to scholarly communication is what he calls the editorial signature: a named editor, embedded in a research community and backed by an institution, accepting personal accountability for a published claim under conditions of genuine uncertainty. Registration, archiving, dissemination, and reward allocation all have dedicated institutions handling them - arXiv, repositories, citation platforms, funding bodies. Certification of research findings, discoveries, and hypotheses is what journals do and what no other institution replicates in the same way. The AI-born architecture is organised around exactly that observation: AI absorbs the surrounding infrastructure, and the signature stays with the editor.

The component technologies for this design already exist in working pilots at eLife, Octopus, aiXiv, and Manubot, each demonstrating a different piece of it. Floridi ends with a concrete proposal: to pilot the AI-born architecture at "Philosophy & Technology," the Springer-Nature journal he edits, over the next twenty-four months, with results reported in the journal itself.

At Cellformatica we spend our days helping researchers prepare their manuscripts for submission, so we do not need to squint at the horizon to see this coming. The journals that move first will lead this transition. The ones that wait will discover that waiting was itself a decision.

‍

[Floridi, "The editor's signature: a proposal for AI-born journals," SSRN 2025. Posted: 24 Apr 2026 - https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6642959]

The article is prepared by the head of Cellformatica content team

‍