At KPG, we approach AI with genuine enthusiasm paired with hard-won skepticism—a pattern I suspect is representative of many mid-sized Asian publishers today: eager to use AI to remove friction from repetitive tasks, but unwilling to let it anywhere near the judgmentcalls that define quality, voice, and accuracy.

In practice, this means AI never touches a first draft unsupervised. For translation, we have never used AI to generate a starting draft; it is used only as a secondary check, run alongside a human translation for comparison. We apply the same principle elsewhere: drafting and translating English-language correspondence, processing data, compiling our internal catalog, brainstorming ideas, and producing rough cover-concept drafts that sharpen the brief we give to designers. In every case, AI produces raw material; an editor then edits, corrects, and aligns it, since left unsupervised it can drift, invent, or simply get things wrong.

We have also learned, through direct experience, where AI still falls short. Two examples stand out: converting OCR’d text from scanned PDFs into clean, reflowable Word documents for reprinting older backlist titles originally laid out in now-obsolete software (not InDesign), and building book indexes—both still require substantial manual correction.

I would welcome the chance to compare notes with other participants on which workflows have genuinely benefited from AI, and which remain, for now, stubbornly and necessarily human.

Indonesia is at a genuine turning point. In 2026, the government tabled a draft amendment to the 2014 Copyright Law that would make Indonesia one of the first countries in Southeast Asia to explicitly legislate AI’s place in copyright through targeted revisions to roughly twenty existing articles alongside entirely new provisions, including a chapter addressing works created with or by artificial intelligence. Notably, the draft would grant copyright protection to AI-assisted works only where there is meaningful human contribution, while fully AI-generated works would receive no protection at all, and it would separately ban AI from imitating a creator’s distinctive style. It would also require technology platforms to compensate publishers when their content is aggregated, republished, or used to train AI models, with proceeds distributed through collective-management bodies.

The main risk I see is definitional: the threshold for “human contribution” remains undefined, which leaves publishers and editors in an uncomfortable gray zone precisely where our daily work happens. The main opportunity is that this ambiguity is forcing publishers to articulate, more rigorously than ever, what human editorial judgment actually contributes to a text—a conversation our industry has long taken for granted. For a market like Indonesia’s, with hundreds of small and mid-sized publishers and a strong translation sector, getting this threshold right will determine whether AI becomes a tool that supports local voices or one that quietly erodes their economic and creative footing.