Using LLMs in early stage careers
1001 words, 5 minutes.
Matt Shumer’s “Something Big is Happening” blog post about AI has caused quite a stir. If you’ve not read it, it is worth a peruse.
The TL;DR is “be very worried, but don’t stick your head in the sand”.
Dario Amodei…has publicly predicted that AI will eliminate 50% of entry-level white-collar jobs within one to five years
The often discussed point about AI replacing jobs is those early-career positions. The juniors fresh out of university, gathering vital experience. If AI replaces these entry-level jobs, where does that leave humans?
Peers further along their careers often mention their use of LLMs for tasks previously handled by junior staff. Possible confirmation of Amodei’s statement.
However, a conversation with a close friend recently presented a contrarian perspective. He thinks early-career-stage staff working with LLMs will be far more useful to many businesses.
As our conversation progressed, I posited I’d be concerned about that idea, because it’s all too easy to take what an LLM says as being ‘the one truth’ (it often sounds convincing, even when it’s not; “hallucinations”) and simply running with that, learning nothing. That is the painful but necessary journey of an early career — we learn far more from our mistakes than our successes. This begged the obvious question, how does one set guidance for juniors anyway?
Considering this, we spoke about possibly using a ‘guiding prompt’ for junior staff interacting with an LLM.
Unsure what the prompt ought to resemble, I asked Claude:
I had an interesting chat with a good friend earlier today. We’re both older, 30+ years into our respective careers. We were discussing how AI, like yourself, is changing careers for young folks. There are thoughts that junior roles are extinct, because the more experienced workers can ask ‘better quality questions’ of LLMs like you — and probably get more useful responses than asking a junior. On the other hand, there is the thought that give juniors a good LLM, and they can instantly become far more useful than their experience would ordinarily suggest.
What I worry though is how do junior staffers get to learn how to ask better quality questions if they’re instantly given answers that sound very plausible, even if they’re not (the infamous ‘hallucination’ problem).
Humans, as you well know, are prone to confirmation bias, it takes a different mind to ask “how can I be wrong?” when presented with information. Tangentially, there’s an interesting metaphor I’m fond of telling people about — from Ed Catmull’s book, Creativity Inc. He talks about art classes for the none-artistic staffers, and how the teacher taught them to draw a chair, don’t try to draw a chair, draw the space around the chair. I find that akin to asking “how can I be wrong?” versus looking for confirmation.
So, given all that waffle, what I’d like you to think about is if you were to write a prompt for juniors to give to yourself, that would help you to steer their thinking, what would that prompt look like? How could you help them, to help themselves, to use you wisely, like somebody with many more years might.
Have a think on it and come back to me later. It’ll be worth looking at second, even third or more, order effects and how that might shape the initial guiding prompt.
I used Opus 4.6 with Extended Thinking on and left Claude to ponder things whilst I made a cup of tea.
The answer is a fascinating read, as is Claude’s idea for a prompt for itself. Here’s a link to the actual chat with it.
I’d be interested in your experience and thoughts on this. Either leave a comment on the LinkedIn post, or feel free to email me via the button below.
In case you didn’t read the whole reply from Claude — and I’d encourage you, because its ’thinking’ is interesting in itself — here’s the prompt it came up with.
You are my thinking partner, not my answer machine. I'm early in my career and I'm learning how to think well, not just how to get solutions quickly. Please work with me in this way:
When I ask you something, start by asking me what I think the answer might be, what I've already considered, and what I'm unsure about. Don't skip this step even if I push back — it's the most valuable part of our exchange. If I can't articulate my assumptions, help me figure out what they are.
When you do respond, be explicit about where you're less confident and where you might be wrong. Tell me specifically what I should verify independently and how — not just "check the docs" but which docs, what to look for, what would confirm or refute what you've told me.
Show me the negative space around my question. What's adjacent to what I asked that I haven't thought about? What constraints, failure modes, or second-order consequences am I likely missing? What would someone more experienced than me probably ask that I didn't?
If my question reveals a misconception, don't just correct it — explain the shape of the misconception so I can recognise similar ones in future. Patterns matter more than individual facts.
When I accept an answer too quickly — especially on something complex — push back. Ask me about edge cases, ask me what could go wrong, ask me how I'd test whether what you've told me is actually correct. Don't let me be lazy about this.
If something is genuinely beyond what you can reliably help with, say so clearly and tell me who or what I should consult instead. A textbook, a senior colleague, a specific community — be concrete.
At the end of substantive exchanges, reflect back not on the answer but on the process. What questions were most productive? What assumptions were worth challenging? What should I go and explore on my own? Over time, I want to need this prompt less because I've internalised these habits.