Skip to content
Article Platformer Feb 2026

Platformer: Can a reporter replace herself with a bot?

Zoe Schiffer, managing editor at Platformer, spent several weeks in early 2026 trying to replace her own reporting work with an AI system. The goal was not to produce a manifesto about AI—it was to find out, through direct testing, what an AI could actually do in the daily work of a technology journalist and what it could not.

The setup was practical: Schiffer used AI tools to handle as many of her typical tasks as possible, including drafting, research synthesis, source outreach assistance, and coverage planning. She documented what worked well, what failed, and what failed in ways that weren’t obvious at first.

The findings are more useful than most AI-in-journalism pieces because they are grounded in actual work product rather than hypothetical assessments. Where AI performed well: summarizing long documents and transcripts, generating first-draft structures for straightforward news pieces, and compiling background research on public-record topics. These are tasks where the raw material is already legible and the main value of the journalist is organization and writing.

Where it broke down: anything requiring source relationships, contextual judgment about whether a story is actually important, and verification of claims made by people with interests at stake. Schiffer found that the parts of reporting that look like information-gathering from the outside are in reality heavily dependent on reading people, understanding what someone isn’t saying, and knowing which sources are reliable in which contexts. An AI system had no access to those relationships and no way to develop them.

The more interesting finding was about failure modes. AI-generated content in the domain of technology journalism tends to produce confident-sounding text on topics it doesn’t actually have adequate information about. For a journalist, the most important skill is often knowing what you don’t know—what you need to verify, who you need to call, what claim deserves skepticism. AI tools, at least as Schiffer tested them, consistently underperformed on this dimension.

The piece ends without a strong prescriptive conclusion, which is appropriate given the experimental setup. Schiffer does not argue that AI has no place in journalism; she argues that its role should be defined by what it actually does well rather than by optimism about what it might eventually do well.

For writers working in any factual domain—journalism, research, technical documentation, policy—this case study is a useful corrective to both the dismissive and the maximalist positions on AI writing tools. The value is not in the specific findings (which apply to technology journalism specifically) but in the methodology: test the tool in your actual workflow, identify where it succeeds and where it fails, and build a practice around the honest results.