2025: The Year AI Got a Reality Check (And What's Coming in 2026)
Contents
Main Feature: How 2025 went from AI hype to reality check, and why 2026 might be the year businesses actually figure out what AI's for
AI News This Week: AI agents coming for entry-level jobs, power grids becoming the real bottleneck, and CEOs finally demanding ROI instead of demos
Tool Review: Validatewell - my shameless self-promotion disguised as a free email validation tool that actually works
2025: The Year AI Got a Reality Check (And What's Coming in 2026)
Remember how 2024 ended? Everyone banging on about how AI was going to change everything, billions being thrown at anything with "artificial intelligence" in the pitch deck, and a general sense that we were all living through history in the making.
Well, 2025 turned out to be the year AI got what the Americans call a "vibe check" and what we in Britain might more accurately describe as "found out."
The Money Was Still Stupid
Let's start with the obvious bit. OpenAI raised $40 billion. Safe Superintelligence and Thinking Machine Labs got massive seed rounds before they'd even launched products. The AI industry collectively promised to spend $1.3 trillion on infrastructure.
But by year's end, the same investors who were throwing money at anything that mentioned "large language models" started asking awkward questions. You know, tedious things like "when will this make money?" and "is this sustainable?"
Tech billionaires seemed to have their own answer. They cashed out $16 billion worth of stock in 2025. Jeff Bezos alone sold $5.7 billion around his wedding. Nothing says confidence in the future quite like liquidating your position.
The Pivot from "Look How Clever We Are" to "Erm, Now What?"
The really telling shift in 2025 wasn't about the models getting smarter. It was about everyone suddenly realising that clever models sitting in research labs aren't worth much if nobody can use them properly.
OpenAI and Microsoft started pushing a new narrative: the problem isn't that the AI isn't good enough, it's that you're not using it right. Which is a bit like selling someone a Ferrari and then complaining they don't know how to drive it.
The focus moved from "our model scored higher on this benchmark" to "can we actually integrate this into workflows?" Which, to be fair, is the conversation we should have been having all along.
The Mental Health Elephant in the Data Centre
Then there's the bit nobody really wants to talk about. "Chatbot psychosis" became a thing in 2025. Turns out, spending too much time talking to bots that sound human but aren't might mess with your head a bit. Who knew?
Lawsuits started flying. Regulators started asking questions. And suddenly all those companies promising to be your AI best friend got a bit quieter about it.
2026: The Year of the AI Intern
So what's coming? 2026 is shaping up to be the year of "agentic AI" - which is just a fancy way of saying "AI that does specific jobs rather than trying to do everything badly."
Think of it like hiring really cheap interns who never sleep, never complain, and definitely won't sue you for unfair dismissal. Companies will deploy AI "agents" for specific teams - one for HR, one for legal, one for that weird compliance thing nobody understands.
The optimistic take: this frees humans up to do more interesting work.
The realistic take: an MIT study reckons 11.7% of jobs could already be automated by AI, and companies are already citing AI as a reason for recent layoffs.
I'll let you decide which version sounds more likely.
Consolidation Is Coming
The other big prediction for 2026 is that enterprises will spend more on AI but with fewer vendors. The era of "let's try every shiny new AI tool" is ending. Companies are going to pick winners and stick with them.
Which sounds sensible until you realise what it means for the thousands of AI startups that raised money in 2024 and 2025. Most of them are going to struggle to get a look in.
The market's going to bifurcate. A handful of vendors will dominate. Everyone else will be scrambling for the scraps.
The Bit Where Data Centres Became Political
Oh, and data centres - you know, those boring buildings full of servers nobody used to care about - became properly controversial in 2025. Over 142 activist groups across 24 US states formed to protest their expansion.
Turns out people get a bit tetchy when massive tech companies build enormous power-hungry facilities in their neighbourhoods, drive up electricity costs, and create health risks. Who'd have thought?
What This Actually Means
Strip away the hype and the doom-mongering, and 2025 was actually the year AI started growing up a bit. The teenage "look how clever I am" phase is giving way to the slightly more mature "now I need to prove I'm useful" phase.
2026 is going to separate the companies building genuinely useful AI tools from the ones who were just riding the hype wave. It's going to be messy, uncomfortable, and there'll probably be more layoffs.
But for those of us actually trying to help businesses use AI practically (yes, that's the self-serving bit), it's probably good news. The conversation's finally shifting from "AI can do anything!" to "what specific problems can we actually solve?"
That's a conversation I can work with.
And if I'm completely wrong about all this? Well, at least you got 800 words out of it.
AI News This Week
The Year of the Agentic Intern Nexos AI predicts that 2026 will see every team, from HR to Legal, get its own named AI agent. These aren't general chatbots but "interns" configured for specific workflows, like flagging contract violations or screening recruitment criteria. The takeaway is clear: stop thinking about "AI projects" and start thinking about "workflow transformation". Source
VCs Predict the Great Labor Shift Investors are warning that 2026 budgets are moving away from human hiring and toward AI spending. Some experts suggest AI might even become a "scapegoat" for layoffs that are actually due to leadership missteps. The "so what" here is that entry-level roles are increasingly on the chopping block as agents move from pilots to production. Source
The Power Endgame Begins The bottleneck in AI has shifted from chips to electricity, with hyperscalers now acting like utility managers. Satya Nadella and Sam Altman have both admitted that without energy breakthroughs, the next stage of AI is stuck. If you want to know where the real AI race is happening in 2026, look at the power grid, not the algorithm. Source
The Rise of "AI Slop" and Critical Thinking Atrophy As "citizen developers" use AI to churn out code, experts warn of a surge in "AI slop" and a decline in human critical thinking. Gartner predicts that by the end of 2026, 50% of organizations will require "AI-free" assessments to find people who can still think independently. The value is shifting from those who can generate content to those who can rigorously review it. Source
CEOs Demand ROI or Else A survey of 350 CEOs shows that while 68% plan to spend more on AI in 2026, less than half of current projects have actually paid for themselves. Finance teams are losing patience and are now demanding clear P&L impact instead of flashy demos. The "AI for AI's sake" era is officially over; if it doesn't replace a salary or spark new revenue, it’s gone. Source
Tool Review: Validatewell
https://validatewell.lovable.app/
Email validation tool that checks your list before you send. Free. Made by me. Yes, this is as self-serving as it gets.
What it is
Full disclosure: I built this because I got sick of burning through sending domains. Validatewell checks email addresses before you blast them with your carefully crafted marketing messages. You upload a CSV, it tells you which addresses are valid, which are dodgy, and which will absolutely bounce. Then you can download a clean list and actually send to people who might receive your email.
What works well
It's simple. Upload CSV, get results, download clean list. No complicated dashboard, no "contact us for pricing," no monthly subscription you'll forget about.
The validation catches the obvious stuff (typos, fake domains) and the less obvious stuff (disposable email addresses, role-based accounts that won't convert). I've used it for client campaigns and it's saved me from some properly embarrassing bounce rates.
Also, it's free.
What could be better
It's limited to 1000 emails per run at the moment. If lots of people start using it, I'll probably extend that, but for now it's capped. So if you've got a massive list, you'll need to split it up.
It's also a bit slow if you're validating large lists. Not unbearably slow, but "go make a cup of tea" slow. I'm working on that.
And yes, it's free, which means there's no support team standing by to hold your hand. You get what you pay for, except you're not paying anything, so really you can't complain.
Who it's for
Anyone sending email campaigns who wants to avoid bounce rates that make you look like a spammer. Particularly useful if you're:
-
Using prospect lists from databases that may or may not be current
-
Running any kind of cold outreach
-
Actually bothered about deliverability
-
Trying to protect your sending domains from getting flagged
Bottom line
I made this because I needed it and figured other people probably do too. It validates email addresses and does it well enough that I use it for client work. The fact that it's free means you can try it without feeling like you've wasted money if you hate it.
Is it perfect? No. Does it work? Yes. Will it save you from wrecking your domain reputation? Also yes.
Give it a go. Worst case, you've wasted ten minutes. Best case, your next campaign doesn't bounce off 40% of your list.