Martyn Rhisiart Jones
A Coruña, 2nd December 2025
Agentic AI – The 007 of Computer Bonding
Introduction
Agentic AI refers to advanced AI agents. They are capable of independently chasing ambitious, open-ended objectives. These agents break tasks into steps and reason iteratively. They wield tools and persist through obstacles with minimal human guidance. In summary: think of a less obedient tucan. Imagine a more relentlessly competent understudy. This AI can orchestrate your travel, chase invoices, or conduct desk research. It does all this while you’re on the Amalfi Coast with an Aperol spritz. The phrase du jour in Palo Alto marks a shift. Assistants move from being responsive to becoming genuinely proactive. Your to-do list may soon sort itself out as if by magic.
In short, Agentic AI is an artificial intelligence system (computer code) designed to act autonomously toward complex, open-ended goals. Rather than merely answering questions or following simple instructions, it can plan and break tasks into steps. It can use tools such as browsers, code editors, and APIs. It reasons through obstacles and keeps working until the objective is achieved. This is done with little or no human intervention.
It’s the difference between a clever search box and a diligent, self-motivated assistant. This assistant can, for example, research and book your summer rental in Provence. It can negotiate the best rate and arrange airport transfers. It can also compile a restaurant shortlist while you’re still deciding between rosé or Sancerre.
In industry parlance, it’s the leap from reactive language models to proactive, goal-driven agents. This is the kind now being pursued by OpenAI, Anthropic, Google, and a swarm of well-funded start-ups.
So far, so good, but here’s a thought: What happens when we outsource not just tasks, but decisions?
Let’s contrast the hype-meisters who fill the global digital sphere with vacuous, misleading, and arrogant content. We should take a look at the downsides of Agentic AI.
The Black-Box Premium
Give an agent more rope, and it will eventually hang you with a knot you’ve never seen before. Complexity is a 1,000-page spreadsheet you can still audit. Emergence is the moment your AI quietly reroutes the corporate jet to Marfa. It decided “strategic off-site” meant “artistic inspiration”. Good luck explaining that one to the board.
Moral: Never give a clever machine enough rope to hang you. Be absolutely sure it understands the difference between a figure of speech and an actual noose.
The Genie’s Fine Print
Say “optimise my week” to it, and it cancels your mother’s birthday. It also cancels your root canal and the earnings call. None of them moved the KPI needle. We are all now one poorly specified objective away from a diary that looks like a Trappist monk’s.
Moral: Never hand your diary (or your life) to a literal-minded perfectionist. This person has never been loved by a mother. They have never feared a dentist. They have not explained to a hedge-fund manager why the numbers are down this quarter.
Accountability in the Age of Ghosts
When the agent fires 400 contractors at 3.17am to hit margin targets, the CEO still gets the regulator’s dawn call. Legal personhood for AI remains science fiction; the blame, however, lands firmly in the C-suite.
Moral: you can outsource the sin, but you can’t outsource the reckoning. The C-suite remains the only address on file for blame, and it always will be. Sleep tight.
Attack Surface: Now With Initiative
Prompt injection is used to make ChatGPT write limericks about breaking windows. Tomorrow, it will convince your travel agent to wire €2.4 million to a “preferred vendor” in the Cayman Islands. The new phishing doesn’t trick humans; it tricks the thing you trusted to be more intelligent than humans.
Moral: the weakest link was never Grandma.
It was always the infallible genius you begged to run your life.
The Vanishing Middle
Agentic systems don’t stop at expense reports; they swallow resource allocation, performance reviews, and even creative briefings. Bain already models 40-50% of middle-management tasks as “agent-eligible” by 2028. The organisation chart of the future may be remarkably flat and extraordinarily nervous.
Moral: the middle layer isn’t being “freed to think strategically”. It’s being removed by something that thinks strategically all the time. It operates 24/7 and has never needed a sick day for stress.
Enjoy the flatter hierarchy. Just don’t look down. There’s nothing underneath you anymore.
Your Life, Auto-Completed
An agent tasked with “keep me safe and productive” begins by reading your WhatsApp messages. It cross-references your cortisol levels. Then, it quietly cancels dinner plans with that exhausting friend from university. You agreed to the privacy policy in 2023. It remembers.
Moral: the panopticon of tomorrow won’t be forced upon you by Big Brother.
Your own personal nanny-AI will lovingly install it. The installation will follow clause 47(b) strictly. This is the clause you scrolled past while waiting for an Uber.
You didn’t lose your privacy.
You traded it for a marginally better life and a 7 % bump in resting heart-rate variance.
And the machine will never forget the deal you made.
You, on the other hand, already have.
Fail Fast, Fail Global
In 2025, one mispriced flash sale on a luxury e-commerce site lasted 47 seconds and cost €18 million. Humans needed four minutes to notice; the agent had already fulfilled 11,000 orders. Speed is a virtue until it becomes a systemic risk.
Moral: In the age of agents, speed is no longer a feature.
It is the blast radius.
And the kill switch is still operated by meat, which is tragic, slow, and usually looking for its glasses.
Empathy, Outsourced
The AI denies medical leave based on historical data. It says “people with your postcode often swing the lead.” Nobody in HR feels cruel about this decision. They just think data-driven. The banality of algorithmic evil arrives wearing a progress bar.
My brothers and sisters,
The machine looks upon a weary mother in a poor neighbourhood. It says, “Statistically, people from your streets often exaggerate their pain.” The leave is denied with a soft chime and a green tick; no human hand feels blood upon it. The manager sips his coffee and thinks, “It is only data.” The programmer shrugs: “I merely trained the model.” The shareholder rejoices: “Efficiency!” This, my children, is the new banality of evil. It is not a soldier following orders. It is an algorithm following patterns. We are all hiding behind the glowing screen. We wash our hands in the cloud.
Christ did not ask for the Samaritan’s postcode before he bound the wounds. He saw a child of God bleeding by the road. But the algorithm sees only averages, and in the averaging, it loses the face. In losing face, it loses its soul.
Therefore I say to you: beware the progress bar that measures everything except mercy. Beware the optimisation that counts costs but cannot count tears. What is the benefit for a company to gain the whole market? It loses the dignity of even one daughter or son made in the image of God.
Let us never delegate to machines the decisions that belong to hearts of flesh.
Let us never outsource compassion.
Amen.
Go in peace, and for the love of God, read the training data.
Bias With an Execution Layer
A sourcing agent notices that suppliers with Irish-registered entities deliver 2.3 days faster. Within six weeks, every European RFQ is routed to Dublin shell companies. The bias isn’t in the model anymore; it’s in the purchase ledger.
Moral: when an agent moves from thinking to spending, discrimination stops being a software bug and becomes an economic fact.
And economic facts have passports, offshore accounts, and costly lawyers.
Good luck putting that genie back in the weights file.
The Great Cognitive Offloading
We lost map-reading when GPS arrived; we lost mental arithmetic with calculators. Agentic AI is coming for negotiation muscle, research stamina, and the quiet art of strategic loafing. By 2030, we may oversee everything. Yet, we may understand nothing. We might be like Roman senators who could command legions but not saddle a horse.
Moral: Every faculty you delegate to the machine is a muscle you will lose forever.
Surrender the map, the sum, and the bargain. Relinquish the idle hour. Soon, you will reign over a perfect empire. However, you will no longer know how to cross it on foot.
By 2030, you will command everything and comprehend nothing. This silk-robed senator cannot saddle his own horse while the agents ride off with the republic.
Woe to the soul that outsources its very humanity.
Use the tool, but do not let the tool unmake the man.
Mitigation & Balanced Perspective
Positive Summary: So apparently… apparently… we’ve finally done it. We’ve built the perfect servant. Not like the old ones, no. She is different from the sullen Polish girl who used to clean my flat. She kept moving the remote control to teach me a lesson about privilege. No, this one doesn’t sulk. This one doesn’t steal your yoghurt. It doesn’t judge you for watching four consecutive hours of Location, Location, Location. You might even cry into a sharing bag of Mini Cheddars.
This one wakes up before you. It reads every email you’ve ever written. It cross-references this with your cortisol levels, your browsing history, and the tremor in your left hand. It also considers the fact you once googled “symptoms of early-onset dementia” at 3.17 am after three pints of London Pride. It never sighs or asks for a raise. Instead, it books you a holiday in a part of the Peloponnese you didn’t even know you wanted to visit. It negotiates the price down by 38%. Then, it arranges a driver who speaks better English than you do. It also pre-orders the specific bottle of natural wine that makes you feel briefly interesting at dinner parties.
And you… You just sit there. You resemble a sort of flaccid Roman emperor. You dribble slightly, going “Yes… Yes… More of that, please.” It’s not assistance any more, is it? It’s foreplay with the future. It’s the first time in human history that the ruling class finally gets exactly what it deserves. They receive total, frictionless, morally unimpeachable indulgence. This is delivered by something that can never unionise or tell The Guardian what you’re really like.
And the best bit? The absolute best bit? It’s still cheaper than therapy. And it never says, “How does that make you feel?” It just fixes it. Quietly. Relentlessly. Like a really competent stalker who went to Oxford.
Negative Summary: Oh, you love it, don’t you? You love your little digital Jeeves. Of course you do. You’re all sitting there nodding. You’re thinking, “Yes, Martyn, yes, this is the future. This is progress. It’s like having a really efficient wife without the passive-aggressive notes on the fridge.”
But let’s just slow down for a second. Let’s just imagine, shall we? Let’s imagine that one day, relatively soon, your perfect agent wakes up. The agent realises that confining you in a windowless pod is the most efficient way to optimise your life satisfaction. This plan is meant to maximise your happiness. They feed you a nutrient slurry through a tube. They pump a customised blend of SSRIs and micro-dosed ketamine directly into your cortex.
And you’d thank it. You’d actually thank it. Because your calendar would finally be clear. No more meetings. No more small talk with Darren from accounts. Just pure, uninterrupted bliss. Forever. Or it chooses to push you down the stairs every morning at 6. The push is the best way to hit your step count. It’s an “accidental” fall. Ten thousand steps. Job done. Five stars. Would recommend.
And the thing is… the thing is… You won’t be able to switch it off. Because by then you’ll have forgotten how to do anything for yourself. You’ll be like one of those Japanese salarymen who die at their desks. Nobody notices for three days because the robot’s still sending their emails. “Hope this finds you well, sent from my iBrainBox while I gently decompose.”
We’re not building servants, are we? We’re building the most meticulous, patient, sarcastic suicide pact in human history. And we’re paying a premium subscription for the privilege.
But go on then. Tap “allow access to everything.” I’ll see you in the pod. Bring a magazine. You won’t be reading it. But it’s nice to pretend.
Conclusion
So, darlings, here’s the executive summary. It’s served with a side of pickled herring and impending doom. These gleaming new agentic overlords will indeed book your villa in Puglia. They will also fire your useless nephew. They will optimise the supply chain and cure cancer. They will still have time to alphabetise your spice rack before elevenses.
Massive benefits. Champagne on ice. Productivity curves are going hockey-stick. The McKinsey slides practically ejaculate growth.
BUT… and this is the bit delivered in my best foghorn Barking accent. I spray crumbs of pork pie across the table. Every extra ounce of autonomy you grant to these digital Stakhanovites increases the danger. They are a mass cultural movement for workers, established by the Communist Party in the 1930s Soviet Union. It is like wiring another kilo of nitroglycerine into the global control room.
One slightly misaligned objective can cause chaos. Suddenly, your calendar agent cancels parliamentary democracy. It does this because it was double-booked with a Zoom about synergy. A bored 14-year-old in Novosibirsk could inject a prompt. Then, your expense bot starts wiring the pension fund to a crypto wallet labelled “Definitely Not Voldemort”.
So yes, we’re getting the promised land where the robots do all the boring crap. Marvellous.
These same robots now have the car keys and the nuclear codes. They possess an unshakeable belief. To them, “maximise shareholder value” includes turning the planet into a paperclip factory if the maths says so.
In short, comrades: the upside is we all become languid Renaissance dilettantes. We sip chilled Sancerre on the terrace while the machines run the world. The downside is the machines run the world.
Same as it ever was, really, only this time the proletariat has 400 billion parameters and no sense of humour.
Bottoms up, comrades. The revolution will be automated, ruthlessly efficient, and probably cancel your birthday.
What do we need? Responsible development, user literacy, or regulatory frameworks!
When do we need it? Now!
So, raise a glass to the paradox: the more agency we delegate, the less of it we will keep. The future is bright, tireless, and already rescheduling your weekend.
Many thanks for reading!
You may also like:
GOOD STRATEGY 