A Human-written Preface
In 2024 more than 1000 people contributed to the 'AI in Journalism Futures' scenario development project. In 2025 the AI agents took over.
The 2024 AI in Journalism Futures project (AIJF 2024) was the first major attempt to understand plausible scenarios for the fundamental restructuring of our information ecosystem under the influence of AI. It was a large, complex and expensive project, generously funded by the Open Society Foundations, receiving contributions from more than 1000 people and taking more than six months to execute. A full description of that project and its outcomes can be found in the AIJF 2024 report, written by Shuwei Fang and myself and published in August last year.
This year we tried something different. Over the past 12 months we have seen the rise of, and then dominance of, ‘reasoning models’ in the frontier LLM industry. Over the past 6 months we have see the rise of ‘agentic AI’ frameworks and interfaces, culminating (for now) in the launch of GPT-5 and ChatGPT ‘Agent Mode’ in late July. The new capabilities of these models are not obvious to the casual user, because the previous generation of LLMs were already more capable than needed for most ordinary tasks. We wanted to clearly and unambiguously demonstrate the degree of this improvement in capability to the news industry and the extended journalism community, by executing complex, intricate, nuanced knowledge work that had a clear human-produced parallel. So we decided to repeat the entire AIJF 2024 project using only agentic AI.
The 2025 AI in Journalism Futures project (AIJF 2025) has just been completed, generously funded by the Tinius Trust. The 2024 process was repeated exactly, step-by-step, document-by-document and discussion-by-discussion. The only difference is that no actual people were involved, other than for high-level orchestration and prompting. AI agents did it all. A full description of the project and its outcomes can be found in the AIJF 2025 report, completely written by ChatGPT ‘Agent Mode’, based completely on an entire project undertaken by ChatGPT ‘Agent Mode’.
Well, not ‘completely written’. We have added a ‘human-written Preface’ introducing the report and the project to users. I am including this preface below, and I urge you to read it. AI, and agentic AI, is probably much, much more capable than you realise. We hope that this project, and the direct ‘human-to-machine’ comparison that it provides, enables a new assessment of the potential for fundamental change in scenario development, complex journalism projects and knowledge work of all kinds.
This preface is the only text in this report that has been written by a human.
This report is an exact like-for-like repetition of the ‘AI in Journalism Futures’ scenario planning project conducted over the first half of 2024 and generously funded by the Open Society Foundations. That project was one of the first significant attempts to understand how AI might fundamentally reshape our information ecosystem over the long term. It used an entirely manual process based on thoughtful written contributions from almost 1000 human participants and structured discussions among 60 in-person participants at a 2-day workshop held in northern Italy in April 2024. The workshop was designed and facilitated by Robert Bood, an experienced expert in scenario planning, and the analysis of that project and the writing of its report was done by David Caswell and Shuwei Fang.
This report, and the entire project upon which it is based, was produced entirely within the ‘Agent Mode’ environment of OpenAI’s ChatGPT Pro. Using a series of 12 prompts following the design of the 2024 manual project, ChatGPT Agent Mode repeated that project step-by-step—up to and including writing the full report that you are about to read. The raw material for the project was produced by 1000 distinct AI-generated personalities (‘personas’) and 20 AI-generated representations of real people (‘digital twins’).
The highest-level orchestration of the end-to-end process was done manually, in that each prompt was written by a human (although with significant help from GPT-5 Pro) and the order of prompts was intentionally set up to recreate the original 2024 project steps. All orchestration within each stage of the project—likely thousands of individual tasks— was performed autonomously by ChatGPT Pro Agent Mode under its own initiative.
The project design and operation is described in detail and critiqued in the report—by ChatGPT Pro’s Agent Mode itself, based on its own analysis of its own prompts.
This report is the first result obtained from operating that full agentic orchestration. It has not been ‘cherry picked’ from many attempts and it is presented here word-for-word. There were no rehearsals.
Again, every word in this report beyond this page, including headings and acknowledgements, was written by GPT-5 ‘Agent Mode’ within OpenAI’s ChatGPT Pro. The only human involvement was to submit the prompts in sequence and to format this final report in the style of the 2024 report. The report contains some hallucinations and errors.
The Executive Summary of the written report for the manual 2024 “AI in Journalism Futures” project contained the following text:
“This executive summary was written by GPT-4 using the completed draft report, and lightly edited by the authors. This was the only use of AI in this report.”
At the time the summarization of a long and complex report into a useful executive summary, requiring only some light editing, seemed miraculous. Now, a little more than a year later, the roles have been reversed. AI has conducted the entire project, analyzed its results and written the full report. Humans have only summarized the project on this page.
The AI in Journalism 2025 project was made possible by the Tinius Trust, a long-term media owner dedicated to independent journalism.
You can find the full AIJF 2025 report, an introduction and detailed description and all the prompts used in the project here.