# The One-Button Wager: Who Wins the AI Tug-of-War in Medical Communications > Pharma is betting that one-button AI authoring tools can replace medical communications work end-to-end. Our thesis: that wager misreads what the work is — and the tools that win will be the ones that side with the writer. Mantir's manifesto for the co-creation era. > There is a wager being placed across the pharmaceutical industry right now: that the work of medical communications — the decks, the monographs, the response documents, the careful sentences that carry a molecule's story to the world — can be generated by pressing a button. Money is moving accordingly: sponsors are funding and buying [one-button authoring](/perspectives/one-button-authoring/) tools, and Eli Lilly's participation in Revisto's $4M seed round (Business Wire, Nov 2024) is the visible tip of a broad pattern. **We are taking the other side of that wager.** Not against AI — we build AI — but against the theory of the work that the button implies. ## The wager, stated fairly The automation case is not stupid, and pretending otherwise loses the argument. Deliverables in this industry are templated. Source material is finite and structured. Output language is constrained by regulation. If any intelligence work should be automatable end-to-end, it's work where the inputs are enumerable and the outputs follow rules. So the wager goes: buy the button, generate the deliverable, let review catch what's wrong, and let the agency line item go to zero. Every knowledge industry is hearing a version of this pitch right now. And in every one of them, the same discovery keeps repeating: **the deliverable was never the work.** The work is the ten thousand judgments that the deliverable is merely the residue of — which claim can survive review, which figure means what it appears to mean, which reviewer will read which sentence which way, what the client said in March that makes this slide a landmine. People who do intelligence work know this. People who buy it at portfolio distance often don't — until the output meets reality. ## Where the button breaks In medical communications, "reality" has a name and an address: it's called review. Every generated deliverable must pass through the [MLR wall](/perspectives/the-mlr-wall/) — committees and signatories who verify, claim by claim, that every statement is substantiated by evidence, at the page-column-paragraph level. And here the automation model hits its structural problem, the [substantiation gap](/perspectives/substantiation-gap/): **generation produces sentences; review demands provenance.** A model can write "reduced LDL-C by 42%" in a perfect house style. It cannot make that number be in the source. The gap between fluent and *substantiated* is the entire regulatory risk of the industry — and one-button tools, by producing more fluent-but-unverified content faster, don't close the gap. They industrialize it. This is the [verification asymmetry](/perspectives/verification-asymmetry/): generating claims is now nearly free; verifying them still costs everything it always cost. When generation gets cheap, verification becomes the bottleneck, the cost center, and therefore the point of control. **Whoever owns verification owns the workflow.** ## The other side of the wager We build for the opposite theory: the [co-creation model](/perspectives/co-creation-model/). Not AI instead of the writer — AI as the writer's instrument, pointed at exactly the parts of the work that were never the craft: hunting for the substantiating passage, rebuilding reference packs, re-annotating decks, checking for drift, pre-running the review. The writer keeps the judgments; the machine keeps the ledger. Where one-button tools treat the writer's expertise as the thing to eliminate, [writer-side AI](/perspectives/writer-side-ai/) treats it as the thing to arm — because in a regulated industry, the expensive resource isn't prose. It's *defensible* prose. We didn't arrive at this theory in the abstract. We built Claims hand-in-hand with medical writers who have decades in this work, and every feature is shaped like the work actually is: claims deep-linked to evidence at writing time, annotations in the format reviewers really read, an agentic pre-review that catches overreach before a committee bills a round for it. Sourced-only, by construction — the tool cannot cite what isn't in the references. ## How we think it plays out The tug-of-war won't end with one side's flag in the mud — see our [field guide to 2030](/perspectives/field-guide-2030/) for the long version. Sponsors will keep buying buttons; some tasks genuinely will collapse into them; and [agency compression](/perspectives/agency-compression/) is real and already underway. But the compression squeezes toward exactly the thing buttons can't do: judgment under regulatory accountability. The medcomms professionals who thrive won't be the ones who out-type the machine or the ones who surrender to [human-as-the-loop](/perspectives/human-as-the-loop/) rubber-stamping. They'll be the ones armed with tools that make their judgment faster, their substantiation bulletproof, and their work verifiable at a glance. That's the side we've picked. The button is betting the writer is overhead. We're betting the writer is the product. --- Source: https://medcom.claims/perspectives/the-one-button-wager/ · Updated 2026-07-13 · © Mantir, Inc.