For the past two years, AI output in the workplace has almost always taken the same form: paragraphs in a chat column, then copied to Word or Excel, and then tidied up manually. Now, the format has changed. What comes out is no longer text, but finished products: well-organized documents, tables with functioning formulas, small pages with clickable buttons.
Anthropic calls it an artifact. Technically, it appears in a separate window next to the conversation and can be modified, saved, or shared without having to copy its contents again. This feature is available in all packages, including the free one.
This change in format may seem cosmetic. In reality, it is not.
Finished products demand a different way of checking
If the output is a paragraph, the way to check it is by reading. If the output is a budget calculator, reading is not enough. You have to try it out.
The difference is evident at the workstation. A projection table may look convincing until you fill the volume column with zero and the result remains positive. A dashboard can be neat until you change the date range and the numbers do not move at all. A form can look beautiful until someone types a period as a thousand separator.
Three inexpensive tests that almost always find something: input extreme values, leave one mandatory field blank, and then change one assumption and see if the numbers below change. The third test often fails because models frequently write calculation results as static numbers, not as formulas.
Those who rarely use it have not reached this stage
The PwC Global Workforce Hopes and Fears 2025 survey, which involved 49,843 workers in 48 countries including 812 respondents from Indonesia, shows a significant gap. In Indonesia, 69 percent of workers reported having used AI in the past year, but only 16 percent use it daily.
The gap between "ever" and "daily" correlates with perceived benefits. Daily users in Indonesia report a productivity increase of 96 percent, while infrequent users report only 75 percent. For job security, the figures are 82 percent compared to 63 percent.
Pete Brown, Global Workforce Leader at PwC, states that training alone is not enough. According to him, the jobs themselves need to be redesigned, and the division of roles between humans and machines needs to be redefined.
Domestic policy direction is moving in that direction. After an evaluation by Bappenas, the Ministry of Communication and Digital has shifted the focus of digital literacy programs from device introduction to upskilling and AI competencies.
What does not change hands
One thing remains in human hands: where the numbers come from. Artifacts can tidy up, recalculate, and present. They do not know whether the sales data you pasted has been finalized or is still last week's draft version.
Therefore, the most valuable checks are not about appearance, but about the source of the materials and assumptions. If a projection uses a growth rate of 12 percent per month, the question is not whether the formula is correct, but who set that 12 percent and on what basis.
Testing the behavior of an artifact takes five to ten minutes. Correcting decisions already made based on incorrect numbers takes much longer.
Sources
- PwC Indonesia – AI adoption is boosting productivity, particularly among Indonesia's Gen Z, yet skill gaps across generations remain a challenge, press release February 23, 2026 → pwc.com/id
- Anthropic Help Center – What are artifacts and how do I use them? → support.claude.com
- Warta Ekonomi – Komdigi Geser Fokus Literasi Digital ke Upskilling dan AI → wartaekonomi.co.id
- ANTARA News – Wamenkomdigi ungkap literasi digital cara baru penting hadapi era AI → antaranews.com