AI Just Got a Promotion. It Doesn't Just Recommend Anymore. It Decides.
AI Just Got a Promotion. It Doesn’t Just Recommend Anymore. It Decides.
By Akshay A. Walimbe
There was a time and it wasn’t that long ago when the worst thing an algorithm could do to you was recommend a bad movie on Netflix.
You’d scroll past it. Maybe roll your eyes. Move on with your life. The machine suggested, and you decided. That was the deal. The AI played advisor, and you held the remote.
That deal is over.
Something happened in the last two years that most people haven’t fully processed yet. AI got a promotion. It went from the assistant who hands you options to the manager who makes the call. And it happened so quietly that you probably didn’t notice the handover.
From Suggestions to Decisions
Think about how you interacted with AI five years ago. Spotify suggested a playlist. Google Maps offered a route. Amazon said “customers who bought this also bought that.” You were always in the loop. You always had the final say.
Now think about what AI does today.
An AI system reviews your resume, scores it, and rejects it without a human ever reading your name. A lending algorithm evaluates your creditworthiness and denies your loan application before any loan officer knows you exist. A trading bot executes thousands of transactions per second, moving real money through real markets, faster than any human could blink, let alone approve.
The AI didn’t just get smarter. It got authority.
There’s a term for this new breed of AI: agentic AI. These aren’t chatbots that respond when you type a question. These are systems that perceive their environment, plan a series of actions, execute those actions, check the results, and adjust all without waiting for you to say “go ahead.”
The old AI was a search engine. The new AI is an employee with a company credit card and no supervisor.
The Workday Problem
Let me tell you about a case that should make every hiring manager and job seeker sit up straight.
Workday one of the world’s largest HR software platforms, used by thousands of companies to screen job applicants is facing a class action discrimination lawsuit in the United States. The allegation? That Workday’s AI powered hiring tools systematically discriminated against applicants based on race, age, and disability.
Here’s what makes this case different from the usual “algorithm gone wrong” stories. A federal judge, Rita Lin, ruled that Workday’s software “is not simply implementing in a rote way the criteria that employers set forth, but is instead participating in the decision making process by recommending some candidates to move forward and rejecting others.”
Read that again. The judge said the AI isn’t just following orders. It’s making decisions.
The plaintiff, Derek Mobley a Black man over 40 with anxiety and depression applied to over 100 positions at companies using Workday’s platform. He was rejected by every single one. Not by a human who read his resume. By a system that scored him and moved on to the next applicant in milliseconds.
Now, I’m not saying every rejection was wrong. But here’s the question that should bother you: if a human hiring manager rejected 100 applications from the same person, at some point someone would ask questions. When a machine does it, nobody notices. There’s no one to ask. There’s no face across the table.
India Is Already in the Deep End
If you’re thinking this is an American problem, let me bring it closer to home.
India’s financial system is already running on autonomous AI. On the National Stock Exchange, algorithmic trading now accounts for roughly 73% of stock futures volume, according to NSE data for FY2026. These aren’t assisted trades where a human clicks “confirm.” These are autonomous agents executing at speeds measured in microseconds, making decisions about real money in real markets.
In banking, the RBI’s FREE AI Committee survey of 612 regulated entities found that 20.8% are now using or developing AI. The Reserve Bank Innovation Hub developed MuleHunter.AI an AI/ML powered system that detects and flags suspicious mule accounts. According to an RTI response from December 2025, 23 banks have now implemented it. UPI exceeded 18 billion monthly transactions in Q4 2025, and NPCI has deployed AI driven fraud detection that operates in real time early trials with four banks cut false positives by 28% while improving detection accuracy, according to MediaNama.
Then there’s Telangana’s Samagra Vedika. This algorithmic system was designed to detect welfare fraud. According to Amnesty International’s April 2024 assessment, it cancelled over 1.86 million food security cards and rejected more than 142,000 fresh applications between 2014 and 2019 many of them wrongful exclusions caused by faulty data and bad algorithmic decisions. People lost access to food rations because a machine decided they were ineligible, and nobody told them why.
Amnesty International’s assessment was blunt: “Automated decision making systems such as Samagra Vedika are opaque, and they flatten people’s lives by reducing them to numbers using artificial intelligence and algorithms.”
That’s not a recommendation engine suggesting a playlist. That’s an autonomous system deciding who eats and who doesn’t.
The Speed Problem
Here’s what makes agentic AI fundamentally different from the AI we’re used to: speed.
When AI recommends and a human decides, there’s a buffer. A pause. A moment where someone can say, “Wait, that doesn’t look right.” Agentic AI removes that buffer. The system perceives, decides, and acts often in the time it takes you to read this sentence.
In May 2010, the US stock market experienced what’s now called the “Flash Crash.” In 36 minutes, nearly a trillion dollars in market value vanished. The Dow Jones dropped almost 1,000 points. It recovered most of it within minutes but the damage was real. Automated trading systems had triggered a cascade of sell orders, each reacting to the other’s behavior, spiraling downward at machine speed.
No human caused it. No human could have stopped it in time.
That was 2010. The algorithms today are orders of magnitude more sophisticated, faster, and more autonomous. And they’re not just in stock markets anymore. They’re booking your flights, processing your insurance claims, evaluating your job applications, and approving or denying your loans.
Pine Labs in India recently partnered with OpenAI to embed autonomous AI directly into payment infrastructure agents that can negotiate supplier terms and manage payments without human intervention. Cashfree Payments launched embedded checkout directly into conversational AI interfaces. The frontier is here, and it’s moving fast.
The Accountability Vacuum
Now here’s the part that should really concern you.
When a human makes a bad decision denies your loan unfairly, rejects your application illegally, cancels your food ration card wrongly you have recourse. You can challenge it. You can appeal. You can look that person in the eye and say, “Explain yourself.”
When an AI agent makes the same decision, who do you look at?
The developer who built the model? The company that deployed it? The cloud provider that hosted it? The data vendor that supplied the training set? India’s AI Governance Guidelines, released by MeitY on November 5, 2025, propose a “graded accountability” framework but the guidelines are voluntary. Not enforceable. Not law.
The AI (Ethics and Accountability) Bill, a Private Member’s Bill introduced in the Lok Sabha on December 17, 2025 by BJP MP Bharti Pardhi, proposes penalties of up to five crore rupees. Compare that to India’s own data protection law, the DPDPA, which allows penalties up to 250 crore. Or the EU AI Act, which can fine companies up to 35 million euros or 7% of global turnover. It is worth noting that as a Private Member’s Bill, its chances of becoming law in its current form are slim only 14 such bills have become law since independence, the last in 1970. But it signals the direction of legislative thinking.
Five crore for an AI that wrongly denies thousands of people food rations. That’s the current state of accountability in India.
And here’s the structural gap that nobody is talking about: what happens when two AI agents interact with each other? When your insurance company’s AI negotiates with the hospital’s AI, or when a trading algorithm interacts with a counterparty’s algorithm and the interaction produces harm that neither system was individually designed to cause?
As MediaNama’s February 2026 analysis put it: “When autonomous agents unfairly deny a borrower a loan or shut a merchant out of a marketplace, responsibility splinters across developers, deployers, cloud providers, and integrators, where each may be compliant; the harm arises from their interaction.”
Nobody planned the harm. Nobody approved the harm. But the harm happened. And the law doesn’t know where to point.
The Question Nobody Is Answering
AI agents are already making decisions that affect your money, your career, your access to government services, and your daily life. They’re getting faster, more autonomous, and more deeply embedded in systems you depend on.
The governance frameworks? Largely voluntary though SEBI’s binding algorithmic trading regulation, mandatory from April 2026, shows India can move to enforceable rules when it chooses to. The penalties proposed in the AI Bill? Pocket change for the companies deploying these systems. The legal frameworks for agent to agent interactions? They don’t exist yet in India or anywhere else.
We gave the machine a promotion. We gave it decision making authority. But we forgot to give it a performance review. We forgot to build the escalation path. We forgot to ask the most basic question of management:
When the machine makes the decision, who do you hold responsible?
I’m have written a book about exactly this how AI and automated systems make decisions about your life, where accountability disappears, and what we can do about it. If you want to know morea about this book or order a copy, you can do it here: https://akshaywalimbe.com/beyond-bias/