The evolving economics of AI, Part 1
AI has claimed more than 100,000 human jobs in the first six months of 2026. Now, it's faltering at being human at the workplace.
Artificial Intelligence is typically sold as infinitely scalable digital labour—faster, more productive, and much cheaper than humans. The logic is compelling. Unlike human employees, AI systems incur no recruitment, training, benefits, office, or payroll costs, yet can draft emails, write code, analyse documents, and provide customer support around the clock. Nor do AI agents suffer from fatigue, endless meetings, coordination bottlenecks, or office politics.
That promise has encouraged companies, including some of the world’s largest tech firms, to eliminate thousands of jobs while betting heavily on autonomous AI agents capable of planning, reasoning, and executing tasks. The industry has shed over 100,000 jobs in first half of 2026 amid AI-led cost cuts.
However, latest developments show that the economics of AI is now being re-examined—not only in terms of cost, but also effectiveness.
In this week’s edition of Tech Talk:
- Does RBI's AI risk framework account for Agentic AI?
- Claude Mythos, Fable restored as US lifts curbs
- Senior leadership at India Inc. gets new AI scorecards
- AI Tool of the Week: ChatGPT Pronunciation Guidance
Human > AI
That Ford Motor Co. decided to rehired 350 veteran engineers after AI-driven quality control systems fell short of expectations underscores the continuing importance of human judgement. Despite deploying AI extensively to improve efficiency and reduce costs, the American carmaker concluded that automated systems could not replicate the practical expertise of human engineers.
This is no isolated case. The likes of Klarna, Nubank, and UPS have demonstrated that AI can deliver substantial savings in structured domains—customer support, logistics, route optimisation, etc.—but the economics changes when tasks require judgement, contextual understanding, or emotional nuance. Klarna ultimately reintroduced human agents to address customer dissatisfaction.
Even some of AI’s strongest advocates are confronting these trade-offs.
- Microsoft Corp., one of the largest investors in OpenAI and Anthropic, has reportedly cancelled internal licences for Anthropic’s Claude Code.
- An Nvidia Corp. executive recently acknowledged that, in some cases, the cost of running AI systems now exceeds that of employees using them.
- McDonald’s reassessed aspects of AI deployments after discovering that large-scale implementation can become unexpectedly expensive.
- Starbucks recently scrapped an AI-powered inventory system after employees reported incorrect counts and product mislabelling.
AI can fail, spectacularly
A February 2026 paper titled Agents of Chaos has documented serious reliability and safety concerns with AI Agents, including leakage of sensitive information, execution of harmful commands, false claims of task completion, and compliance with unauthorised users.
Then there is the cost of the infrastructure itself. The world’s largest technology companies are spending hundreds of billions of dollars on AI data centres, chips, energy, and water resources to sustain increasingly capable models.
High performance comes at higher cost
That Anthropic PBC has sharply increased prices for premium Claude services used by finance professionals signals that while basic AI may eventually get commoditised, high-performance AI for finance, law, software engineering, and enterprise workflows may end up resembling premium infrastructure rather than cheap digital labour.
AI, thus, may not eliminate the cost of human work so much as redistribute it, replacing payroll expenses with infrastructure, supervision, and the potentially enormous cost of machine error.
The importance of being human
Human skills remain the most sought-after capabilities among employers globally, yet shortages in these skills are as severe as—and often more severe than—gaps in technical expertise, according to the QS Global Employer Survey. The QS World Future Skills Index 2027 finds that higher education systems are struggling to keep pace with a labour market being reshaped by AI.
The report argues that bridging this gap will require more than curriculum reform, calling instead for a broader reset of higher education policy, funding, governance, institutional agility, and employer investment in lifelong learning.
India highlights both the challenge and the opportunity. It ranks fifth globally in Future of Work readiness (96.0) but only 18th in Skills Alignment (82.7), underscoring the mismatch between labour market transformation and the ability to produce job-ready graduates. As one of the world’s largest higher education systems, India’s success in closing this gap will shape not only its own economic future but also the global supply of skilled talent, the report says.
The AI capability unlocked today is ChatGPT Pronunciation Guidance
What problem does it solve? Most business professionals anglicise unfamiliar names or quietly avoid them, because there was no reliable way to hear the correct pronunciation before walking into a room. Before a client call in Seoul or Warsaw, they guess the client’s name. Before a pitch in Tokyo, they skip a term because they have never heard it said aloud.
ChatGPT now fixes this. Ask how to pronounce any word in over 60 languages and you get a written phonetic breakdown and a playable audio example in one response.
How to access: https://chatgpt.com
ChatGPT Pronunciation Guidance can help you:
- Sound client-ready: Get the right pronunciation of a client name, city, or brand before any international call.
- Nail the unfamiliar: Handle foreign-language terms from a new market confidently, without mispronouncing in front of a room.
- Build as you go: Get phonetic text and audio together so the correct form sticks before you walk in.
Example: A sales director is preparing a pitch to a pharma company in Tokyo. Her slide includes three Japanese executive names she has never said aloud.
- Prompt the names: Ask “How do I pronounce Yoshida, Nakamura, and Takahashi? Include phonetic breakdown”, and ChatGPT returns text and playable audio.
- Request audio: Click to play the pronunciation directly in chat, no separate app needed.
- Add context: Ask “Correct pronunciation of Takeda and Eisai in a Japanese business setting” for domain guidance.
- Specify register: Add “Use formal Japanese conventions” and the output adjusts to match a professional register.
- Switch languages: Move to “How do I say these city names in Arabic for my next call?” without starting over.
- Save to prep notes: Copy the phonetic text into your meeting notes before you walk in.
What makes ChatGPT Pronunciation Guidance special?
- Audio and text together: Unlike dedicated pronunciation tools, ChatGPT gives both, right in the chat where you are already working.
- 60+ languages without switching apps: Mandarin, Arabic, Polish and more, covering the range global professionals actually encounter.
- No setup required: No plugin, no special mode. Just ask in any regular ChatGPT chat.
Note: The tools and analysis featured in this section demonstrated clear value based on our internal testing. Our recommendations are entirely independent and not influenced by the tool creators.
AI BITS & BYTES
Does RBI’s AI framework account for Agentic AI?
The Reserve Bank of India’s recently released draft guidance on an AI framework for banks deserves credit for addressing limitations ranging from hallucinations to model drift and adversarial attacks.The key proposals include strict validation for third-party models and mandatory “kill switch” mechanisms for AI.
That said, the next generation of AI systems may not behave like models at all. The Agentic AI systems can plan, make decisions, interact with external tools, execute actions and coordinate with other agents with limited human intervention.
In the context of banking, AI Agents may eventually approve loans, execute financial transactions, communicate with customers and trigger downstream actions autonomously. At that point, the question facing regulators changes fundamentally. The challenge is no longer supervising models that make predictions, but governing systems that make decisions and act on them.
Anthropic restores Claude Mythos, Fable
On Tuesday, Anthropic said that the US Commerce Department has lifted export controls on its most advanced AI models—Fable and Mythos. “We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5,” Anthropic said in a post on X.
Anthropic had abruptly disabled its Mythos 5 and Fable 5 models following the export-control order on 12 June, over national security risks.
“We’ll begin restoring access tomorrow, and will share an update soon,” the company said. Anthropic also thanked users for their support. Read more.
Senior leadership gets new AI scorecards
Employers investing in AI-led upskilling programmes for their staff want their money’s worth. And the sure-shot way they are doing it is by including AI adoption as a parameter in performance appraisals.
From banking to consulting, business houses have started rating 5-15% of their senior leadership based on how they used AI beneficially, along with factors like people management, operations and financials. Read more.
Why an AI correction can be India’s opportunity
Big Tech remains publicly confident that artificial intelligence (AI) will eventually justify the huge sums of capital pouring into the sector. Yet, concerns have grown over less-than-commensurate returns and high market volatility amid reports that OpenAI may defer its long-awaited public offer of equity.
All this suggests that the breakneck pace of today’s AI boom may not be sustainable. Goldman Sachs projects capital spending of $5.3 trillion by hyperscalers between 2025 and 2030, much of it in AI chips and data centres.
Should today’s AI narrative falter or investment plans go awry, though, it may prompt investors to look anew at AI latecomers like India. Read more.