There's no such thing as an entry-level AI role. We established that last month. And yet here we are with a grant (two actually) to help entry-level AI hiring in the South West and Wales, by March.
Surely AI has to be making some new entry-level roles, not just replacing them all. (We can hope.)
Oddly, this is full circle for me. Back in 2018 I started researching who gets left out of AI careers and why. One of the things I found then - pre-ChatGPT, pre this whole wave of generative AI - was that when people thought about AI jobs, they thought about coding. Building AI. Data scientists, machine learning engineers, people with PhDs writing code for neural networks. People building robots.
There wasn't anyone else - even though back then, plenty of people were working on ethics, business adoption, all kinds of sides of AI.
That hasn't really changed. What has changed is who's affected. Back then it was mostly physical workers being displaced - robots on factory floors, AI in warehouses. Now it's knowledge workers. Lawyers, analysts, marketers, writers, accountants. The people who thought they were safe.
And while we know nobody is actually hiring an AI role - they're hiring AI Product Managers, Forward Deployed Engineers, AI Strategists, Data Analysts / Marketeers / Ops / [insert office based job here] who use AI - we do think these roles can be split into three categories - Using AI, Building with AI, Building AI. And all three have an entry point. They just look very different from each other.
Using AI
This is the majority of AI employment. Right now, today. Using AI tools - Copilot, ChatGPT, Claude, whatever's built into the software you already use - to do your job better. I think AI will become like spreadsheets - an expected baseline skill for anyone in an office. We wrote about this a while back.
PwC's 2026 AI Jobs Barometer found UK AI user roles grew 65.8% in 2025 - driven by people using AI within existing roles, not building it. The EU AI Act went further: AI literacy is now a legal requirement for any company using AI, with enforcement started this August.
There is a tension here though. Our research found that most of the hiring in this category is currently experienced people with AI skills - brought in to replace the juniors, not alongside them. That's not everyone. And we think it's the wrong call - teach people the domain, give them real tasks, let them use AI safely and figure it out. That's how you build the seniors of the future.
Entry level? Yes - and it should be.
Building With AI
This one has changed the most in the last few years. And is the most interesting to me.
Coding is no longer a requirement to build. That is the bigger shift than people being able to write emails faster.
People are creating AI-powered products, automations and workflows without needing to write code (or understand what they are doing - though we don't really recommend that). There are two versions of this: absolutely no code, using a platform like Lovable and building by prompting; or using existing AI models like OpenAI or Anthropic in an application.
The second is actually what Amy and I do with the tbh platform. We haven't built the AI algorithms behind Dotly - we've architected the platform cleverly around them. That distinction matters. You don't need to understand how a model is trained to build something genuinely useful with it. We can code, and we do read code, but Claude definitely helps us out. We're also massive fans of using Lovable to build with AI, or automation tools like Make or Zapier to build amazing workflows.
The new AI and Automation Practitioner Level 4 Apprenticeship (levy-funded, new cohorts from May 2026) literally says no coding required in the standard. A government-approved, Ofsted-assessed qualification built around no-code AI building. We hope this catches on - it feels like a big signal.
Entry level? Yes - and more accessible than most people realise. Juniors really add value here. We have a junior working as an Automation Engineer 50% of their week - and it's great.
Building AI
Training and fine-tuning models. Data scientists, ML researchers, MLOps engineers.
These roles typically require a relevant degree or prior experience. They are less likely to be the open, learn-on-the-job entry points that the other two tiers can be. That doesn't mean they're impossible to break into - but the path in looks different, and it's a different conversation from what we're focused on with this programme.
Why Does This Matter?
We're thinking about this wrong - who cares if it's formally an AI role. Most roles will use AI, and we can all agree on a few things: entry-level roles are at risk, and we need to act. Hiring is broken in the age of AI.
My PhD found that when you asked people about AI jobs, they assumed you needed coding, maths, a STEM degree. With some reframing - just explaining that most AI roles aren't that - everyone was interested. The framing is the barrier, not the ability.
Maybe now's the time? If you're thinking about how you can support juniors in the age of AI - and how you can derisk hiring and be more supportive - we'd love to chat.