Anthropic is riding the disruption wave. Elon Musk has, contentiously, even called the company “Misanthropic”, suggesting that the company’s AI models are against society and humankind.
Disrupting IT services
In February 2025, Anthropic unveiled Claude Cowork, its agentic AI tool, which has 11 plug-ins that enables it to perform workflows end-to-end, including reviewing contracts, dealing with legal documents, writing code, and coordinating tasks across teams, with no supervision. The market response was brutal, shaving off more than $285 billion in market value across technology and IT services firms, affecting Salesforce, Intuit, Adobe, and Thomson Reuters, among others. Indian IT majors Infosys, Wipro, HCL Tech, and TCS lost between 15 and 21% by way of market capitalisation.
Anthropic’s legal tool is integrated with its own model. It has also released similar tools for sales, marketing, finance, and accounting—signalling that cloud-native software and agentic AI can displace expensive SaaS (software-as-a-service) and cloud options, apart from shrinking manual effort and errors substantially.
Mainframe susceptibility
Then came a blog post, less than a fortnight later, about Claude being able to comprehend and contemporise COBOL—a very old mainframe programming language developed in 1959, which is still dominant in government, airline, telecom, and banking systems. That single post swiftly eroded $30 billion of IBM’s market value and Indian IT stocks took a further knock, with the Nifty IT index dropping almost 5%. COBOL programmers are scarce now, and Claude being able to manage codebases, figure dependencies, rewrite code, and produce documentation, is indeed a game changer.
The inference that IT services companies’ revenues deriving from related consulting, maintenance, and upgrade would get adversely impacted isn’t flawed. The math is indeed revealing—we have 800 billion lines of COBOL code operating worldwide, and the modernising costs would shrink from about $12 to $3 or less per line. As is known, there are surrounding systems, regulatory patches galore, local innovations, and the app layers on top, to consider too. That begs a question—what about the other legacy codebases?
Cyber liabilities
Earlier this month, Anthropic announced Mythos, which had the unprecedented capability to autonomously identify, at scale, hitherto undiscovered cybersecurity vulnerabilities. What created panic in the ranks was when Mythos identified critical vulnerabilities undiscovered for 27 years. The model’s capabilities extend to every major operating system, every major web browser, all widely used software, and IT infrastructure.
Wisely, Anthropic has proactively reached out to government and key industry stakeholders to discuss the model’s defensive and offensive capabilities. Under a cybersecurity initiative called Project Glasswing, they are now restricting access to 40 tech companies, infrastructure players, and security organisations, including Microsoft, Amazon, Nvidia, Cisco, and Google. The aim is to allow rapid patching of the newly discovered vulnerabilities. US Treasury Secretary Scott Bessant and Federal Reserve Chair Jerome Powell convened an urgent meeting of bank CEOs and finance chieftains that was attended by CEOs of Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs, to ensure the risks are understood and acted upon with urgency.
The move emphasises that frontier AI models could accelerate cyberattacks if deployed without adequate precautions. And equally, that AI systems meant to protect can become powerful tools for hackers, with the line between offence and defence blurring. With Anthropic researchers having declared that the model is an order of magnitude faster than previous tools, the time between discovering security bugs and their exploitation is dangerously narrow.
What next?
Claude’s stack demonstrates remarkable capability convergence, a distinction that’s reshaping the tech world. For instance, while robotic process automation requires rule-based scripts, Claude figures it out with adaptive automation. It addresses fragmented workflows with end-to-end automation, builds trust for regulated use with reasoning transparency, and simplifies services by unifying coding, operations, and support.
Consulting could meet its disruptor in Claude. Market analysis, presentation decks, operational diagnostics, and scenario analyses are fair game, challenging slideware consulting and dated generic frameworks. The science of management itself could change in time, with Claude arguably doing a better 24×7 job of decision-making, resource allocation, performance monitoring, and optimisation, with managers becoming AI agent supervisors.
Anthropic’s business has accelerated, as has the demand for Claude, moving from $9 billion revenue in 2025 to an estimated $30 billion in 2026, going by their run rate. Anthropic, currently using tensor processing units from Google, and Amazon’s chips, is even contemplating designing its own chips, as part of its public commitment to strengthen US computing infrastructure.
A lot has changed in an AI-native world. Work is not executed any more by layers of people and tonnes of software. Traditional SaaS interfaces and large teams, internal or outsourced, are passé. What’s needed is a cognitive layer that embeds intelligence into workflows, and orchestrates decisioning and execution end-to-end, that redefines speed, productivity, cost, and competitive advantage. Anthropic is positioned well to play a pivotal role, especially in mission-critical AI systems.
Threats
OpenAI’s Sam Altman has warned that as AI systems move towards “superintelligence”, threats would include biological hazards and mass surveillance, if not controlled. Arguably, the bigger threat is public anger against corporate leaders over societal impact, including large-scale displacement, loss of jobs, and access to healthcare, amidst galloping costs—as exemplified by a series of unsavoury events over past months.
When unknown persons hurled a “Molotov cocktail” into Sam Altman’s San Francisco residence recently, it was interpreted as an illegal expression of hostility to AI’s progress. Is that the greater disruptor to worry about in the near term?