The industry knows the direction. The blockage is people.
That is the blunt result of the Future Newsrooms Study 2026, from FT Strategies and WAN-IFRA, with support from Arc XP, released in June at the World News Congress in Marseille. The survey ran from 19 March to 24 April. It drew on 448 responses from newsrooms in 86 countries. Leaders can describe the destination: engagement, clearer signals of trust, AI, a closer audience, new formats. Execution is something else.
On AI, the study finds the tool is still mostly an efficiency device. Time saved is the most common success metric, cited by 42 percent of newsrooms. The largest barriers are not technical. Skills gaps come first, at 61 percent, then cultural resistance at 52 percent, then unclear use cases at 45 percent. Press Gazette’s write-up adds the texture from the same research: more than half of newsrooms have no AI expert inside editorial, and the most common use is still transcription and translation, at 78 percent. Even there, agentic use is uncommon. About 10 percent say an autonomous system handles that work, and that is the high-water mark for agentic AI in the study.
This rhymes with what UK journalists reported last autumn. Individuals use the tools. The institution has not decided what the tools are for, and it has not trained people to a shared standard. A vision written outside the newsroom, the study argues, is part of the resistance. Journalists adopt what they helped specify.
The fix is unglamorous and local. Put someone in the newsroom whose job is the editorial use, not the vendor demo. Pick use cases a desk can reject or keep. Teach the people who will sign the copy. Stop scoring the programme only in minutes saved, or you will get faster production of the same stories and call it transformation.
Agents will matter to the audience, as other work this year has shown. Inside most newsrooms, in the summer of 2026, AI is still an assistant waiting for an editor. That is a healthy constraint. It is also a training problem, and training problems do not solve themselves.