Big Tech cut grad hiring by 25%. Banks are slashing junior roles. The question every founder now faces is when do you hire a human, and when do you just hand it to AI?
Great data points here, Martin. That 25% drop in grad hiring is brutal.
Juniors doing automatable work are the first casualties. Which is exactly why AI literacy has become non-negotiable at every level.
For juniors: AI skills are now table stakes. Not to replace thinking, but to amplify it. The ones who survive won't be competing with AI: they'll be leveraging it to punch above their weight class.
For seniors: Not everyone needs to become a prompt engineer. But the best leaders? They understand AI capabilities well enough to orchestrate it strategically.
It's not about seniors learning to prompt better. It's about knowing what to delegate to AI vs humans, and how to supervise both.
Yes to all of this. The brutal part is the silent rewiring of what “junior” even means now. You’re not judged on potential anymore. You’re judged on leverage. And if you can't direct AI to multiply your value, you're functionally under qualified, no matter your degree. This is a major shift in how orgs will need allocate trust and resources.
You painted the picture so well. AI skeptics might push back on every word here, even if it’s all true.
The more I build with AI, the clearer it becomes: we’re already living in a time when most jobs will require some level of AI fluency. It’s not a distant future, it’s arriving, whether we like it or not.
Instead of resisting it, I’ve found it far more meaningful to lean in, stay curious, and grow with it.
Thanks Jenny and yes i expect to get some push back but I truly believe what I wrote. You're approach is spot on. The mindset shift isn’t “will this affect me?” It’s “how do I stay relevant when it already has?
I agree with most of your math. I also agree with your comments on the future role (or not) of juniors, but I believe the savings might not be as great as outlined. Why? because LLMs get things wrong. I propose cost saved should be based on the following formula:
I’m not claiming this formula is perfect (it needs refinement), but it’s a starting point.
To explain using a simple example. Let’s assume a solicitor asks their junior to perform some research. The solicitor intends to use the research as part of their response to a client.
The junior uses an LLM rather than their usual laborious manual process and returns the result to the solicitor for checking.
Before AI-driven tools became available, the junior spent 6 hours performing research and the solicitor spent 1 hour checking the work.
Now, using AI tools, the junior spends 1 hour on research, but as LLMs are known to get things wrong, the solicitor spends an extra 1 hour checking.
Assume the junior hourly rate is £20/hour and the solicitor £70/hour. Then the value (saving) of the LLM is £30 (100-70), not £100 (5x20)
Here’s an interesting thought. Before AI (at 6 hours x £70) it made no sense for the solicitor to perform the research, but what if the business dispenses with the junior?
Then, with the solicitor performing the research task, assume there is no additional checking time (because they check as they go). The total cost to the business for the complete research task is now £140 (70 task+70 check), a saving of £50. But the solicitor is a finite resource!
Thanks Phil this is a great addition. Your formula brilliantly captures the hidden cost everyone's missing, verification overhead. We're essentially trading junior capacity for senior capacity and senior time is always the bottleneck. However once teams develop better prompting skills and error-spotting patterns, doesn't that checking time decrease significantly?
I agree, better prompting (context) will certainly help (a saving), but that requires upskilling the junior (cost). To prompt (provide context) you need to understand the subject. It's probable that will require a higher level of experience/skill.
This is me, tech founder with design founder and business founder partners. We don't need to hire to grow. Also it's moving from taste on problem delegation towards taste on problem parallelization; getting the timing right on concurrent delegations moving through one human expert management pipeline is the new challenge. Even more so when you can attack the "AI gets it wrong" problem with just more parallel attempts.
“Avoid Junior Roles” is crazy advice to give a recent college graduate. I worry that we are shutting off the pipeline to create more seniors in the future.
Great data points here, Martin. That 25% drop in grad hiring is brutal.
Juniors doing automatable work are the first casualties. Which is exactly why AI literacy has become non-negotiable at every level.
For juniors: AI skills are now table stakes. Not to replace thinking, but to amplify it. The ones who survive won't be competing with AI: they'll be leveraging it to punch above their weight class.
For seniors: Not everyone needs to become a prompt engineer. But the best leaders? They understand AI capabilities well enough to orchestrate it strategically.
It's not about seniors learning to prompt better. It's about knowing what to delegate to AI vs humans, and how to supervise both.
Yes to all of this. The brutal part is the silent rewiring of what “junior” even means now. You’re not judged on potential anymore. You’re judged on leverage. And if you can't direct AI to multiply your value, you're functionally under qualified, no matter your degree. This is a major shift in how orgs will need allocate trust and resources.
You painted the picture so well. AI skeptics might push back on every word here, even if it’s all true.
The more I build with AI, the clearer it becomes: we’re already living in a time when most jobs will require some level of AI fluency. It’s not a distant future, it’s arriving, whether we like it or not.
Instead of resisting it, I’ve found it far more meaningful to lean in, stay curious, and grow with it.
Thanks Jenny and yes i expect to get some push back but I truly believe what I wrote. You're approach is spot on. The mindset shift isn’t “will this affect me?” It’s “how do I stay relevant when it already has?
I agree with most of your math. I also agree with your comments on the future role (or not) of juniors, but I believe the savings might not be as great as outlined. Why? because LLMs get things wrong. I propose cost saved should be based on the following formula:
(Time saved × hourly rate) – (Additional checking time × hourly rate)
I’m not claiming this formula is perfect (it needs refinement), but it’s a starting point.
To explain using a simple example. Let’s assume a solicitor asks their junior to perform some research. The solicitor intends to use the research as part of their response to a client.
The junior uses an LLM rather than their usual laborious manual process and returns the result to the solicitor for checking.
Before AI-driven tools became available, the junior spent 6 hours performing research and the solicitor spent 1 hour checking the work.
Now, using AI tools, the junior spends 1 hour on research, but as LLMs are known to get things wrong, the solicitor spends an extra 1 hour checking.
Assume the junior hourly rate is £20/hour and the solicitor £70/hour. Then the value (saving) of the LLM is £30 (100-70), not £100 (5x20)
Here’s an interesting thought. Before AI (at 6 hours x £70) it made no sense for the solicitor to perform the research, but what if the business dispenses with the junior?
Then, with the solicitor performing the research task, assume there is no additional checking time (because they check as they go). The total cost to the business for the complete research task is now £140 (70 task+70 check), a saving of £50. But the solicitor is a finite resource!
Thanks Phil this is a great addition. Your formula brilliantly captures the hidden cost everyone's missing, verification overhead. We're essentially trading junior capacity for senior capacity and senior time is always the bottleneck. However once teams develop better prompting skills and error-spotting patterns, doesn't that checking time decrease significantly?
I agree, better prompting (context) will certainly help (a saving), but that requires upskilling the junior (cost). To prompt (provide context) you need to understand the subject. It's probable that will require a higher level of experience/skill.
Thought-provoking stuff Martin! It does play on my mind that all seniors were once juniors in their careers at some point.
It's all well and good for current seniors. But where will the next set of seniors come from?
This is me, tech founder with design founder and business founder partners. We don't need to hire to grow. Also it's moving from taste on problem delegation towards taste on problem parallelization; getting the timing right on concurrent delegations moving through one human expert management pipeline is the new challenge. Even more so when you can attack the "AI gets it wrong" problem with just more parallel attempts.
“Avoid Junior Roles” is crazy advice to give a recent college graduate. I worry that we are shutting off the pipeline to create more seniors in the future.