Media › Harald's Curious Corner
In conversationHarald Overaa hosts Harald's Curious Corner. This conversation was published on 2026-10-20. What was said is below in Harald’s questions and the answers to them, with every quotation checked word for word against the recording’s transcript. The transcript itself is not published here, and the reason is at the foot of the page.
Written from the transcript below and from nothing else.
Harald Overaa came to the London launch of SuperSkills, and he opens by asking about timing: why this book, and why now. Rahim goes back to Shanghai in 2013 and 2014, building EtonX, when an edtech startup showed him machine learning applied to how people learn and it frightened him. The ChatGPT moment excited him instead, and watching his own children use AI at school and university is what made him write the book. What scared him most was technology starting to do the learning for us.
The middle of the conversation is drift versus design. Rahim says there is no line you cross from designing with AI to drifting with it, and admits he drifts himself, which a friend pointed out is the conundrum. Harald brings in social media, built to keep people engaged, and they agree the question is what a tool is for.
Then discernment, which Rahim calls the last mile of thinking: the part AI cannot walk for you. He argues it is trainable, the way taste is trained through comparison, and that the most important AI skill of the next decade is the confidence to say no to a machine and give the reason.
The close is for learning and development teams. Harald suggests they should be architects of capability rather than content creators. Rahim takes the image further, from which walls are load-bearing to what an organisation keeps custody of as human, and warns against installing escalators where people need to climb.
Spoken on the day and quoted as spoken. Where a line rests on something this research has checked, the research page is linked beside it.
“It's the final stretch. It's deciding, it's making that decision. It is that thing that AI can't walk for you”
Rahim Hirji on Harald's Curious Corner. The evidence is on /research/what-is-judgement.
“There's no border crossing between when you're in design and when you're drifting.”
Rahim Hirji on Harald's Curious Corner. The evidence is on /research/design-versus-drift.
“I do think it is trainable”
Rahim Hirji on Harald's Curious Corner. The evidence is on /research/how-is-judgement-trained.
“I think what scared me the most, right, is that technology started doing the learning for us.”
Rahim Hirji on Harald's Curious Corner. The evidence is on /research/does-using-ai-stop-you-learning.
“I am not going to use this today because it's gonna atrophy my judgment”
Rahim Hirji on Harald's Curious Corner. The evidence is on /research/the-use-or-keep-test.
“what do we have custody of from a human perspective”
Rahim Hirji on Harald's Curious Corner. The evidence is on /research/what-is-custody.
Harald’s questions and the answers given, shortened and not rewritten.
A few moments. In Shanghai in 2013 and 2014, building EtonX, I was shown machine learning applied to how people learn and it scared me. The ChatGPT moment excited me. Then I watched my own children using AI at school and university. What scared me most was technology starting to do the learning for us.
You usually don't. There is no border crossing between designing with AI and drifting with it. It is an unconscious step, and I drift myself.
Discernment is the last mile of thinking: the final stretch, deciding, the part AI cannot walk for you. I think it is trainable, in the same way taste is built through comparison.
Being able to say no to a machine with confidence, and give the reason: I disagree with this because of this, or I am not going to use this today because it would atrophy my judgement.
Act as architects rather than content creators. Decide what to build and what not to build, which walls are load-bearing, and what the organisation keeps custody of as human, rather than outsourcing everything to AI.
The other conversations on this site carry the whole thing as text. This one does not, and the reason is worth stating.
The only transcript available for this episode is the automatic one, which mishears names, infers sentence boundaries and supplies no punctuation. It can be cleaned into something readable, and what comes out is a reconstruction rather than a record. Publishing it as the conversation would be a claim this site is not in a position to make. The quotations above are checked word for word against it before this page is built; they are quoted because they survive that check. For the rest, go to the episode and hear it said.
Published here 2026-09-27. The episode was published by Harald Overaa on 2026-10-20. Anything said in a conversation is speech, dated and attributed, and is not restated as a claim anywhere else on this site. Keynotes · Advisory · Boards