Josh Kim speaks at AMWA

AMWA 2025 Recap: Why AI-Generated First Drafts Aren’t the Hard Part

At AMWA 2025, our co-founder and CEO Josh Kim led a session that sparked one of the most engaged conversations of the conference.

The topic?
A perspective that may sound counterintuitive in a world obsessed with AI content generation:

AI-generated first drafts don’t actually matter that much.

Not because they aren’t useful—but because they’re not where most of the real work (or pain) lives.

First Drafts Are Only the Beginning

Over the last few years, the industry has seen an explosion of tools promising to “automate medical writing” by generating first drafts with large language models. And to be clear: first-draft automation is helpful. It can save time, reduce blank-page anxiety, and accelerate early momentum.

But during Josh’s AMWA session, the discussion focused on a more important question:

Where do medical writers actually spend their time?

The answer, echoed by many in the room, was not the first draft.

The Real Time Sinks Come After the Draft

What follows a first draft is where complexity compounds:

• Reconciling content across sections, documents, and modules

• Incorporating reviewer feedback and cross-functional comments

• Ensuring consistency with source documents and prior submissions

• Managing version control, traceability, and QC

• Updating downstream documents when upstream data changes

These are iterative, non-linear, and deeply contextual workflows—and they don’t have a clean, prompt-based solution.

As Josh put it during the session, the challenge isn’t generating text.
It’s everything that happens around the text.

Why This Changes How We Think About AI for Medical Writing

A key theme of the conversation was that AI needs to be contextualized within the full authoring lifecycle, not treated as a standalone drafting tool.

That means:

• Understanding how content is reused, updated, and reviewed over time

• Designing systems that work with messy, real-world life sciences data

• Supporting human-in-the-loop workflows rather than replacing them

• Building for reliability, traceability, and collaboration—not just speed

This is where many AI tools fall short. Large language models are powerful, but they weren’t designed to manage evolving documents, dependencies, and regulatory rigor on their own.

The Problems We Work on Every Day at Artos

At Artos AI, these “after the first draft” challenges are exactly what we focus on.

Instead of optimizing for prompt engineering or one-off generation, we’ve built our platform around:

• Working directly with existing source documents and data

• Supporting iterative review and revision cycles

• Handling cross-document consistency and updates

• Fitting naturally into real medical writing workflows

The goal isn’t to make AI louder—it’s to make it useful where the work actually happens.

Thank You, AMWA

We’re incredibly grateful for the thoughtful, engaged discussion during the session—and especially impressed by the energy in the room at the very end of a packed conference.

Conversations like these reinforce what we believe strongly:
the future of AI in medical writing isn’t about flashy demos—it’s about solving the hard, unglamorous problems that professionals deal with every day.

If you joined us at AMWA 2025, thank you. And if you’re continuing to wrestle with how AI fits into real-world medical writing workflows, we’d love to keep the conversation going.