Anyone can say, in 2026, that human skills matter in the age of AI. The interesting question is who was saying it before it was safe to. Box of Amazing has published an annual reading of emerging technology since 2018, and the newsletter itself goes back to January 2017, which is about five years before ChatGPT; by that summer it was already on its twenty-third weekly issue. The record matters less for any single forecast than for its through-line. In the December 2018 list, written for 2019, two of the fifty themes were "soft skills will be the X-factor in the workplace" and "there will be Iron Men", the idea of humans working alongside machines rather than being replaced by them. That is the SuperSkills thesis and the centaur model of Human at the Start, dated to 2018. This page keeps the record honest, including the parts that aged badly, because a real track record has both.
The answer, in one line
Since 2017. Rahim Hirji has written Box of Amazing, a weekly reading of emerging technology and human capability, since 2017, and has published an annual predictions or themes piece every year since 2018.
The record#
The annual pieces, in order. The early lists were a curated reading of the whole field's predictions, gathered, sifted and given Rahim's own selection and emphasis, rather than a set of personal forecasts. The questions and the later calls are his own. All were published under Box of Amazing, which ran on Mailchimp from 2017, cross-posted to Medium's Predict publication in the early years, moved through Revue, and has been on Substack since 2023.
- 2018: "Predictions for 2018" (January 2018), the first of the annual pieces.
- 2019: "50 Emerging Technology Themes to watch out for in 2019" (December 2018). This is the load-bearing one: soft skills as the workplace X-factor, "Iron Men", AI getting humanised, AI reshaping the jobs market, and AI ending the exam in schools all appear here.
- 2020 and 2021: "50 Emerging Technology Themes to watch out for" continued, published in Medium's Predict. By the 2021 edition Rahim noted it was "the third year" of the list.
- 2022: "50 Emerging (Technology) Themes to watch out for in 2022", and a companion "Resolutions, Reviews and Predictions".
- 2023: "2023 Trends, Plans and Predictions".
- 2025: "20 Questions for 2025" (December 2024), a shift to his own questions rather than a curated list.
- 2026: "20 Questions for 2026" (December 2025), and the call that "2026 will be the year human skills decide who thrives with AI".
The through-line: human capability, early#
Read the 2019 list now and the selection is the tell. Out of fifty themes drawn from across the industry, the ones Rahim chose to elevate include soft skills as the deciding factor at work (theme 17), "there will be Iron Men" (theme 48), "AI gets humanised" (theme 45), "AI, robots and AI robots shake up the jobs market" (theme 12), and "the beginning of the end of exams", AI in schools (theme 39). These are not random. They are the seeds of everything that became SuperSkills: that as machines absorb the tasks, the human capabilities become the differentiator; that the future is human-plus-machine, not human-versus-machine; and that education and the workplace both have to be redesigned around it. The vocabulary came later. The through-line was there in 2018.
How to read a track record honestly#
Two honest caveats, because a record with no caveats is marketing. First, the annual "50 themes" pieces were curation, a synthesis of what the whole field was predicting, sifted and prioritised, not fifty personal forecasts to be graded. Their value is the selection and the continuity, not a hit rate. Second, the lists contain plenty that fizzled or was mistimed alongside the calls that proved right, and both stay on the record. WFH being named the coming norm in a list written in 2018, before anyone had heard of a pandemic, reads well now; other themes did not survive contact with reality. Keeping the misses is the point. It is what makes the hits, and the early focus on human capability, credible rather than convenient.
A living record#
This page is maintained as the running index of the annual reading, and it grows each December. The original issues live in the Box of Amazing archive. The purpose here is not to relitigate old lists but to make one thing checkable: the questions this work now answers in depth, about judgement, capability, learning and where humans belong alongside AI, are questions it has been asking, in public and on the record, since before the current wave began.
What would prove the through-line wrong#
A dated record of what you got right is worth very little on its own, because anybody publishing fifty themes a year will be right about some of them. What makes a track record readable is stating, in advance, what would count as being wrong. G13 of the September 2026 site review asked for that and it is a fair ask.
The through-line is one claim rather than fifty: that AI reaches judgement before it reaches jobs, and that the capability an organisation loses is lost by drift rather than by decision. Three findings would damage it, and none of them is unlikely enough to be a rhetorical device.
- Organisations that automate heavily turn out to retain unaided capability at the same rate as those that do not. The decay literature is the load-bearing evidence here and it is measured in individuals rather than in organisations. If an organisation-level study found no difference, the capability-debt argument would be in serious trouble.
- Human oversight arrangements are shown to catch wrong outputs at a useful rate in ordinary operating conditions. The current evidence is that inserting a person at the end of a process is not, by itself, a safeguard. A large field study finding otherwise would undo most of what these pages argue about oversight.
- The entry-level route into senior work reconstitutes itself somewhere else. The missing rungs argument assumes the rungs are removed rather than moved. If the junior work reappears in a different shape and people still arrive at seniority having done it, the argument was about a transition rather than a loss.
Each December the record above is revisited in public, alongside the annual reading and, from its first edition, the Drift Index. Anything on this page that the year disproved stays on it, with the correction dated, in the same way as everything else in this research. What was wrong is logged on the accuracy page.
Related SuperSkills research#
The through-line runs into the current work: human skills in the age of AI, AI and human judgement, Human at the Start and drift versus design. The full dated record of the newsletter behind these, from January 2017, is at the timeline. The claims themselves, banded by evidence strength and including what remains unknown, are in what we actually know about AI and human capability. For how the discourse itself changed between 2023 and 2026, and why the founding estimates are still quoted over the measurements that complicate them, see the best writing on AI.
About this research#
Rahim Hirji is the author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company. He has written Box of Amazing, a weekly reading of technology and human capability, since January 2017, for more than 25,000 readers. Dates and pieces on this page are drawn from that published record; the early annual lists were curated syntheses of the field, and are described as such. This is a living reference, updated each year.
Reference · SS-2026-012
Hirji, R. (2026). The predictions record. The SuperSkills evidence base, SS-2026-012. https://thesuperskills.com/research/predictions. Last reviewed 25 August 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work