A week before OpenAI announced the appointment of Arvind KC as its chief people officer on Wednesday, Boris Cherny, creator of Claude Code at Anthropic, told Y Combinator’s Lightcone podcast (tinyurl.com/5n8wud6d): ‘Today, coding is practically solved for me,’ and predicted that we will start to see the ‘software engineer’ title ‘go away’. Last May, Dario Amodei had already predicted that 50% of jobs will go away.

Arvind takes on his new role as a key interface between human and cutting-edge AI in the same month when India’s bellwether IT names sold off hard amid a fresh wave of ‘AI will eat IT’ anxiety. TCS hit a 52-week low in a sharp drop, with the broader Nifty IT pack also sliding, and significant value wiped out across the index constituents.

US-listed IT-services names such as Accenture and Cognizant also fell on the same ‘AI-led fears’ narrative, as did any stock which had ‘software’ in its offerings. These fears may be hyped, but they aren’t totally irrational.

• AI agents are real, no longer just slideware Agents are turning into semi-autonomous doers, with OpenAI positioning its coding agent Codex explicitly as a ‘software engineering agent’ that can run tasks in parallel. If coding is democratised and AI agents are doing the heavy lifting, why should an enterprise pay for a 500-person offshore team? The arbitrage business model, predicated on the unlimited availability of cheap human software engineers, is effectively dead.
• Cost curves are taking unanticipated turns One competent engineer, plus a small pack of agents, can produce output that previously required a small team. That doesn’t mean quality magically takes care of itself. But it does mean the labour-to-output ratio that underpins classic IT-services pricing is under pressure.
• Vibe coding Here, you create working software using natural language for prototyping, internal tools, workflows, even customer-facing experiences. Tools like Claude Code and GPT-Codex have done more than just help developers. They’ve democratised the very act of creation. With vibe coding, a product manager with a clear vision can use natural language to describe an application and have an AI agent generate the functional code compressing the path from intent to code.
• Skill disruption isn’t limited to coders Tools are inveigling themselves into analysis, documentation, testing, customer support workflows, and internal ops. Claude’s Cowork, for example, is a collaborative agent positioned as everyday work infrastructure, whether legal or accounting or marketing.

So, it’s not just the market hallucinating. Something structural is happening. However grim it sounds, IT services and the India story built around it won’t die. But the business models and structure, as well as the unit of value, will radically change.

Demand for software is effectively infinite as enterprises keep digitising everything, and the bottleneck is nearly always the IT department. It’s Jevons paradox: if you can remove bottlenecks, you don’t reduce demand but increase it manifold. Every internal process, no matter how niche, will soon be digitised and agentic. Goldman Sachs predicts the US software spend could triple to $2.8 tn by 2037.

What does change, however, is how software is bought and sold. Per-seat SaaS pricing reportedly makes less sense when users include autonomous agents, and the unit becomes ‘tasks completed,’ ‘queries,’ ‘actions,’ ‘tokens,’ and outcomes, not seats. So, SaaS changes from being ‘Software-as-a-Service’ to ‘Service-as-a-Software’. Not selling access to a tool, but delivering an outcome via an agentic layer that lives inside the customer’s workflows. You don’t buy the software, but you buy the work getting done.