Right then. cracks knuckles, stares into the middle distance with the weary optimism of a British person who has just seen the price of a Freddo
You want to know the next digital train to jump on? The one that stops at Platform 9¾, not Clapham Junction? Well, put down that NFT of a pixelated monkey and step away from the “Metaverse” office block. We’ve all been burned before.
We are currently living through the hangover of the “Wild West” AI era. 2025 was apparently the year of “experimentation,” which is a polite way of saying most of us were just feeding the office Excel sheet into ChatGPT and asking it to “make it funnier” (the results were, well, mixed). According to a recent MIT report cited by Forbes, 95% of generative AI pilots were failing in 2025. Ninety-five per cent. That’s a higher failure rate than my attempts to grow an avocado plant from a stone, honestly.
So, as we trudge through 2026, the hype is settling like a fog over a damp moor. The question isn’t “What is shiny?” The question is “What actually works?”
According to Deloitte, Gartner, Forrester, and a bunch of very clever people who talk in T-pot tokens, the train leaving the station right now isn’t a single locomotive. It’s a proper convoy. But if you have to buy one ticket, make it for the Agentic AI Express, with a connecting ticket for Physical AI (the “Bionic Brit” carriage).
Here is the state of play, served with a stiff upper lip and a brolly for the incoming rain of jargon.
1. The Great Correction: Why the Chatbot Era is Dead
We had our fun. We asked LLMs to write wedding speeches and revise contracts. But as the London Business School points out, those models lack context. They are clever generalists who have “read the internet,” but they don’t know your business. They’re like a contestant on University Challenge—brilliant at quizzes, useless at actually making you a cup of tea.
Enter Agentic AI. Gartner lists this as a top trend for 2026. Deloitte calls it “The Agentic Reality Check.” Forget typing a prompt and getting a poem. Agentic AI is about giving an AI a goal—”Book my travel for the Leeds conference, staying under budget, and don’t use Ryanair”—and letting it negotiate with your calendar, the train booking site, and your finance software to actually do it.
It moves from being a tool to a teammate. Or, depending on how the governance goes, a very expensive digital intern who might accidentally order 4,000 stress balls.
2. The Physical Shift (Or, “When Robots Stop Vacing”)
The really juicy bit, however, isn’t just in the cloud. It’s when the AI grows a spine and legs. Physical AI is the convergence of those brainy LLMs with the messy, sticky, physical world.
Forrester’s 2026 report explicitly maps this shift: AI is moving from digital systems into physical settings. We aren’t just talking about the robotic vacuum that still gets stuck under the sofa. We are talking about Vision-Language-Action (VLA) models.
Imagine this: BMW is already using autonomous intra-plant vehicles. Amazon has DeepFleet coordinating robots. But the next step is the humanoid robot. Stanford’s 2026 Tech Review highlights robotics as a frontier technology. The “Momentum 100” list, an AI-generated index from Wikipedia data, ranks reinforcement learning and robotics incredibly high.
Why is this the “train to take”? Because the maths works. Demographics in the West are aging. We have a labour shortage. If you can sell a robot that does the work of a human for $10 an hour with no sick days and no complaints about the temperature of the tea in the canteen, you are going to clean up.
3. The “Infrastructure Reckoning” (The Not-Sexy Bit That Makes You Money)
Now, for the British humour to subside and the serious business hat to go on for a moment. You cannot ride the train if the tracks are made of cheese.
We are hitting a wall called “Inference Economics.”
Sure, the cost of a “token” (the currency of AI thought) is dropping. But because Agentic AI runs in loops—thinking, acting, failing, retrying—your usage goes through the roof. Deloitte warns that cloud-native bills can hit “tens of millions.”
The trend here is Hybrid Computing. The smart money is moving away from putting everything in the public cloud. It’s moving to the Edge. Why send a video of a leaky pipe to the internet to ask “Is this broken?” when a $10 chip on the pipe can tell you instantly?
This is “Geopatration”—moving workloads back from US hyperscalers to regional or national alternatives. For British firms, this isn’t just about speed; it’s about sovereignty and not getting whacked by fluctuating exchange rates.
The Verdict: Should You Queue at the Platform?
So, dear reader, you have three choices of train. Do not miss them.
Train A: The Orchestrator (Agentic AI)
- Why: It fixes the “95% failure rate” problem. It takes AI from a bullshit generator to a process worker.
- The Risk: It requires you to clean up your own data. (I know, I know, that sounds like work.) You can’t have an agent negotiate a workflow if your workflow is “Dave in accounting sends a vaguely worded PDF.”
- British Take: It’s like finally hiring a PA, but one who doesn’t steal your biscuits.
Train B: The Bionic Brit (Physical AI)
- Why: Labour shortages. Logistics. The UK has an aging workforce and a massive productivity problem. Robots that can navigate our cramped, cobbled streets? Gold dust.
- The Risk: The “Sim-to-Real” gap. A robot can learn to stack boxes in a perfect simulation a million times, but the first time it encounters a wet floor sign in a Tesco warehouse, it might have an existential crisis.
- British Take: It’s a hard hat, not a spacesuit. Very practical.
Train C: The Utility Player (Infrastructure)
- Why: The enabler. You might not build the robot, but you can build the chip, the edge computing node, or the “Context Engineering” layer that London Business School talks about, curating the specific data that makes AI useful.
- The Risk: It’s less glamorous. You won’t get on the cover of Wired for selling server racks.
- British Take: The plumber of the digital age. Invisible, vital, and charges like a wounded rhino.
The Final Word
The world is changing fast. Too fast for most of us. The metaverse is a ghost town, crypto is having a mid-life crisis, and the AI hype is deafening.
But the Agentic Physical shift is real. It’s happening in warehouses in Coventry, in operating theatres in Singapore, and in data centres in Iceland. Don’t look for the magic app that solves everything. Look for the agent that automates the boring stuff, or the robot that stacks the shelf, or the chip that powers the brain.
That is the train. It doesn’t look like the Hogwarts Express. It looks like a very efficient, slightly awkward Tube train that runs on time and asks for feedback via a QR code. Buy your ticket, mind the gap between experimentation and execution, and for God’s sake, keep your voice down, some of us are trying to work.
by ROBERT HAMPTON