On 29 November 2022, I published a LinkedIN article called "What is GPT-4 and why should recruiters be excited by it?" Link to article
The next day, OpenAI released ChatGPT.
I'd love to claim I knew. I didn't. It was pure coincidence.
But it means I have something most people talking about AI don't. A timestamped record of what I thought before the hype started.
So I went back and read it. Then I read the follow-up I wrote in 2023, "What is AGI and should recruiters be worried?" Link to article
Most of the direction was right.
What I got badly wrong was the speed. And how far the capability would go.
What I predicted
In 2022, my predictions were modest.
AI would help write job adverts and content. It might "maybe" write a CV coversheet "in the future". It could be built into recruitment databases to suggest better Boolean searches.
In 2023, I went further. I said emails, sourcing, screening, outreach, CV drafting and adverts could all be automated. I gave it 10 years.
At the time, some people thought that was dramatic.
Three years later, much of that list can already be done. Not perfectly, and not without a human checking the output. But it can be done.
(Not every detail aged well. I repeated the rumour that GPT-4 would have 100 trillion parameters. Direction right, specifics wrong.)
Three years, not ten
Job descriptions were the easy prediction. According to SHRM's 2025 research, 51% of organisations now use AI for recruiting, and two thirds of those use it to write job descriptions.
The part I didn't imagine was what came after.
I thought recruiters would use AI. Ask it for a better advert. A faster email. A smarter search string.
I didn't imagine a non-technical recruitment founder building his own AI systems.
I'm not a developer. But over the past year, with AI doing most of the coding, I've built an end-to-end recruitment workflow. It includes:
🎯 A sourcing engine. Give it a brief and it builds the target profile, finds candidates using web scraping and semantic search, scores and ranks them against the brief, enriches their details and hands back a shortlist.
📞 Delivery workflows. Call recordings become ATS notes, candidate follow-ups, CV scorecards, branded coversheets and client-ready reference reports.
🔍 A market engine. It scans who is hiring across our specialisms every week, filters out the noise and identifies the real decision-makers.
A shortlist used to take a full day to put together. It now takes less than an hour.
Remember the coversheet I said AI might write "in the future"? It's now one small output of a workflow that starts before the candidate has even been found.
In 2022, I imagined AI helping me write a Boolean string. In 2026, I describe the problem and a system I built works on it in the background.
That's the gap between what I predicted and what happened.
The real advantage is connecting it all
No single tool is the advantage. Connecting them is.
For twenty years, the ATS was a filing cabinet. Recruiters typed notes into it. Very little useful came back out. Most of it was out of date.
AI-native ATS platforms change that. They're built with AI at the core, not bolted on. And the best ones are open, so your own tools can read from them and write back to them.
That's where it gets interesting.
Your sourcing engine checks who you already know before searching the market. Call recordings update candidate records automatically. Market intelligence flows straight into your client pipeline. Shortlists land in the system, scored and enriched, ready for a human to review.
The ATS stops being a database you maintain. It becomes the brain all your AI tools plug into.
I didn't predict this in 2022. I couldn't have. A small recruitment business wiring its own AI systems into its core database, without an engineering team, would have sounded like science fiction.
Everything is "AI" now
Meanwhile, every recruitment tool is now "AI-powered". A new one launches every week, each with its own monthly subscription.
Here's the reality. A lot of them are a large language model with a nice interface and a price tag on top. A wrapper.
Some add real data, real workflow and real polish. But with a bit of know-how, much of it can now be built yourself. Tailored to how your business actually works. Owned, not rented.
And if every agency subscribes to the same AI sourcing tool, nobody has an advantage. It's just the new baseline.
In 2024 the question was "which AI tool should I buy?"
In 2026 the question is "what can I build that my competitors can't?"
When something costs almost nothing to produce, people produce too much of it. And most of it is bad.
Look at LinkedIn. There's a strong case that content quality has dropped since people started outsourcing their posts to AI. Same structure. Same hooks. Same empty conclusions. People have a name for it now: AI slop.
I should be honest here. In my 2022 article, I proudly revealed that 90% of it had been written by GPT-3.
I wouldn't do that today.
AI is brilliant for getting the creative juices going, challenging your thinking, tightening structure and fixing grammar. I use it for all of those things, including on this article.
But the ideas have to be yours. The opinions have to be yours. The experience has to be yours.
Candidates now use AI to tailor CVs to every job, write cover letters, rehearse interviews and mass-apply.
The AI automates the shortlisting I used to do by hand. Then a human qualifies every candidate on it. A high AI score isn't the same as a good candidate, and I've seen enough of the technology's limits to know it.
AI fatigue is already real. People are tired of AI slop in their feeds, irrelevant automated outreach in their inbox, and AI-generated replies to messages they took time to write.
When everyone automates the conversation, a genuinely human one stands out.
What's left is judgement. Knowing who is genuinely good, not who looks good on paper. Knowing who will actually move, and why. Knowing the market well enough to tell a client their salary is 30% below it. Getting an offer over the line when the counteroffer lands.
In 2023, I wrote that people will always want a trusted human handling their career. I still believe that.
But as everything else becomes automated, commoditised and fakeable, trust stops being a nice-to-have. It becomes the product.