Here is what happened yesterday: Anthropic released the most capable publicly available AI coding model ever built. Claude Opus 4.7 just scored 87.6% on SWE-bench Verified — meaning it can now independently resolve nearly nine in ten real-world GitHub bugs without a human touching the keyboard. In the same quarter this model dropped, 78,557 tech workers globally lost their jobs, and nearly half of those cuts were officially attributed to AI. That context matters, and this article covers both. But first: the model itself. Opus 4.7 processes images at 3.75 megapixels — more than triple the resolution of any previous Claude model — and is available to anyone with a Claude Pro subscription or API key at the same price as what it replaces. Sources: Anthropic official announcement, April 16, 2026; 9to5Mac, April 16, 2026; Axios, April 16, 2026.

This is not a niche technical event. It is a measurable inflection point in what AI can do in the real world — and it lands in the middle of the most turbulent labor market moment the technology sector has seen in a generation. In the first quarter of 2026 alone, 78,557 technology workers were laid off, with more than 47 percent of those cuts attributed by employers to AI and automation — not to market downturns, not to strategic pivots, not to macroeconomic pressure. To AI. In India, TCS cut approximately 20,000 jobs in 2025 — the largest single layoff in the company's history — citing AI-driven restructuring. Indian IT firms filed more WARN notices in Q1 2026 than in all of 2025 combined, signaling that what began as a productivity story is becoming a headcount story. Sources: Nikkei Asia/Tom's Hardware April 2026; Rest of World, February 2026; UnboxFuture, April 2026.

Claude Opus 4.7 did not cause this. The forces reshaping the global IT labor market were already in motion before April 16. But it is the most significant single capability upgrade released into that environment, and it is happening today. This article covers what Opus 4.7 actually does — with specific numbers from official Anthropic benchmarks and independent evaluators — what it means for the US and Indian technology job markets with cited data, what experts say about the trajectory ahead, and the honest answer to the question everyone is actually asking: should you be afraid? Sources: Anthropic official announcement, April 16, 2026; Anthropic Economic Index March 2026 report; multiple cited labor market data sources.

What Claude Opus 4.7 Actually Is — In Plain English

Claude Opus 4.7 is Anthropic's flagship AI model, replacing Claude Opus 4.6 as the most powerful AI the company makes generally available to the public. It launched on April 16, 2026, two months after Opus 4.6 debuted in February — matching Anthropic's recent cadence of major model releases approximately every eight weeks. It is available through the Claude web interface, the Claude iOS and Android apps, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The API model identifier is `claude-opus-4-7`. Source: Anthropic official announcement, April 16, 2026; GitHub Changelog, April 16, 2026.

One important contextual note: Opus 4.7 is not Anthropic's most powerful AI model. That distinction belongs to Claude Mythos Preview — a model Anthropic described as too powerful to release publicly due to cybersecurity concerns, and which is currently being used only by a select group of technology and security companies (Apple, Microsoft, Amazon, Google, and others) as part of Project Glasswing, Anthropic's defensive cybersecurity initiative. Opus 4.7 is the most powerful model available to everyone. Mythos is the model Anthropic has decided the world is not yet ready for. This distinction matters for calibrating what Opus 4.7 represents: the public frontier, not the absolute frontier. Sources: Anthropic Project Glasswing announcement, April 7, 2026; CNBC, April 16, 2026; 9to5Mac, April 16, 2026.

What's Actually New in Claude Opus 4.7: The Five Changes That Matter

Marketing announcements frequently describe incremental improvements as revolutionary. Opus 4.7 has several changes that are genuinely significant — and a few that are primarily relevant to developers integrating the API. Here is what actually changed, sourced directly from Anthropic's official release documentation. Source: Anthropic official announcement and API documentation, April 16, 2026; APIdog technical breakdown, April 16, 2026.

• High-resolution vision (3.75 megapixels): Previous Claude models could process images at a maximum of approximately 1.15 megapixels — a resolution that caused problems with detailed screenshots, documents with small text, complex diagrams, and any visual task requiring precision. Opus 4.7 raises that ceiling to 3.75 megapixels, which is more than triple the previous limit. In practice, this means the model can read fine print in document scans, identify specific UI elements in high-density interfaces, accurately map coordinates in computer-use workflows, and count and measure visual elements with dramatically improved accuracy. Specific gains: low-level perception tasks (pointing, measuring, counting), bounding-box detection, and natural-image localization all show clear improvements. For businesses using Claude for document processing, design review, or computer-use automation, this is the single most impactful change in day-to-day use. Source: Anthropic official announcement; APIdog analysis, April 16, 2026.
• New 'xhigh' effort level: Claude's effort parameter controls how much internal reasoning the model invests before producing output. Opus 4.7 adds an 'xhigh' level above the existing 'high' level — giving developers and users a way to instruct the model to spend significantly more reasoning tokens on hard problems. Anthropic describes this as being designed for 'coding and agentic tasks where quality matters more than latency.' In plain terms: when you're asking Claude to solve a genuinely difficult engineering problem where being right is more important than being fast, xhigh effort is the mode to use. This is directly relevant to the kind of work that software engineers typically spend hours on — the hard problems where a junior developer would get stuck, where a senior developer would need to think carefully, and where previously AI models would produce plausible-looking but incorrect outputs. Source: Anthropic official announcement; APIdog, April 16, 2026.
• Task budgets for agentic workflows (beta): One of the most significant limitations of AI agents running multi-turn tasks has been cost unpredictability. When you set an AI agent working on a complex task — writing and testing a software feature, conducting multi-step research, operating a computer interface to complete a workflow — it can consume an enormous and unpredictable number of tokens before finishing. Task budgets address this by giving Opus 4.7 a token allowance for an entire agentic task, not just a single response. The model sees a running countdown and uses it to prioritize, skip low-value steps, and finish gracefully as the budget is approached. This makes agentic deployments significantly more practical for cost-sensitive enterprise use cases. Minimum task budget is 20,000 tokens. Source: Anthropic API documentation, April 16, 2026; APIdog, April 16, 2026.
• Adaptive thinking (replaces fixed reasoning budgets): Opus 4.6 allowed developers to allocate a specific number of 'thinking tokens' — internal reasoning the model does before producing an answer. Opus 4.7 removes this in favor of adaptive thinking, where the model dynamically decides how much reasoning to invest based on the difficulty of the task. Anthropic's internal evaluations found adaptive thinking consistently outperformed fixed-budget approaches because the model can calibrate reasoning intensity per task rather than applying a uniform budget regardless of complexity. Important technical note: adaptive thinking is off by default and must be explicitly enabled. Source: APIdog, April 16, 2026; Anthropic API documentation.
• Improved cross-session memory: For AI agents handling long-running or multi-session tasks — an agent managing a software project over multiple days, an AI assistant maintaining context about a research initiative, an autonomous worker handling complex enterprise workflows — Opus 4.7 significantly improves the ability to write to and read from file-system-based memory. The model is better at updating notes across sessions, referencing previous context accurately, and moving to new tasks without requiring the user to re-explain what happened before. Anthropic states: 'It remembers important notes across long, multi-session work, and uses them to move on to new tasks that, as a result, need less up-front context.' This is practically significant for any enterprise use case involving sustained autonomous work. Source: Anthropic official announcement, April 16, 2026; 9to5Mac, April 16, 2026.

The Benchmark Numbers: What Opus 4.7 Actually Scores and What That Means

AI benchmark scores are frequently cited without context, which produces both excessive alarm and excessive dismissal. Here is the Opus 4.7 benchmark data sourced directly from Anthropic's official release, with explanation of what each benchmark measures in practical terms. Where competitor scores are cited, they come from Anthropic's comparison table in the official announcement or from Artificial Analysis, the leading independent AI evaluation organization. Sources: Anthropic official announcement, April 16, 2026; OfficeChai benchmark analysis, April 16, 2026; BenchLM.ai, April 16, 2026.

The number that deserves the most attention is SWE-bench Verified at 87.6%. SWE-bench is constructed from real GitHub issues — actual software bugs that developers filed in open-source projects and then resolved. The benchmark tests whether an AI model can independently read the bug report, understand the codebase, write a fix, and produce code that actually passes the project's test suite. At 87.6%, Opus 4.7 resolves nearly nine in ten of these real-world bugs without human assistance. To give this scale: the benchmark was first published in 2024, when the best AI models achieved scores in the low teens. In two years, AI coding capability went from solving roughly 1 in 10 real bugs to solving nearly 9 in 10. The improvement is not incremental. It is a categorical change in what software engineering AI can do. Sources: Anthropic official announcement; SWE-bench leaderboard history; OfficeChai, April 16, 2026.

The Rakuten-SWE-Bench result is also notable because it is not a benchmark — it is a real production engineering environment. Rakuten, the Japanese e-commerce giant, operates a large engineering organization and evaluated Opus 4.7 against Opus 4.6 on its own codebase and engineering tasks. The result: Opus 4.7 resolved three times more production tasks than Opus 4.6. This is not a lab result. It is an enterprise measurement of what the model does when pointed at real software that real engineers are maintaining. Rakuten also reported double-digit gains in code quality and test quality compared to Opus 4.6. Source: Anthropic official announcement quoting Rakuten, April 16, 2026.

The Regression Controversy: Why Opus 4.7 Exists and What It Fixes

To understand Opus 4.7's significance, it helps to understand the context it was released into. In the weeks before the Opus 4.7 announcement, Anthropic faced an unusual wave of public criticism from technical users who reported that Claude Opus 4.6 — the previous flagship model — had gotten measurably worse at complex tasks over time. The criticism was specific and credible. An AMD senior director wrote in a widely shared GitHub post: 'Claude has regressed to the point it cannot be trusted to perform complex engineering.' The complaint was not isolated. Developers building on Claude for production systems reported that the model they had tuned their workflows around was producing worse outputs than it had previously, without any announced change to the model. Source: Axios, April 16, 2026.

The speculation within the AI developer community centered on 'nerfing' — the practice of deliberately reducing a model's capabilities, either to control costs as usage scales, to redirect compute resources to other projects, or for safety reasons that are not publicly disclosed. Anthropic's response was direct: the company denied that any changes to Opus 4.6 were made to redirect computing resources to other projects. Whether other factors contributed to the perceived regression, and whether user perception accurately tracked actual model behavior changes, remains contested. What is not contested is that the technical community's frustration was real, that Anthropic was publicly aware of it, and that Opus 4.7 is the company's direct answer to it. Source: Axios, April 16, 2026.

Anthropic's own framing of Opus 4.7 reflects this context: 'Users report being able to hand off their hardest coding work — the kind that previously needed close supervision — to Opus 4.7 with confidence.' The phrase 'previously needed close supervision' is an implicit acknowledgment that the level of trust required to use AI for the hardest engineering work had not previously been earned, and that the standard is now being raised. The benchmark data supports the claim. Whether it fully resolves the community's confidence concerns will be answered by real-world performance over the next several weeks. Source: Anthropic official announcement, April 16, 2026; 9to5Mac, April 16, 2026.

78,557 Jobs. One Quarter. Here's What the Labor Market Data Actually Shows.

The release of Opus 4.7 did not happen in a vacuum. It is landing in the middle of a technology labor market that is experiencing a structural shift unlike anything in the past decade. The headline number comes from technology layoff tracking and Nikkei Asia's analysis, reported by Tom's Hardware in April 2026: 78,557 technology workers were laid off globally in the first quarter of 2026. Of those cuts, 47.9 percent — approximately 37,638 positions — were attributed by the companies making the cuts to AI and automation: reduced need for human workers due to AI tools, workflow automation, or both. More than 76 percent of the affected positions were located in the United States. Source: Tom's Hardware citing Nikkei Asia, April 2026.

There is an important caveat in interpreting these numbers, and honest coverage requires stating it. OpenAI CEO Sam Altman addressed this directly at the India AI Impact Summit: 'I don't know what the exact percentage is, but there's some AI washing where people are blaming AI for layoffs that they would otherwise do, and then there's some real displacement by AI of different kinds of jobs.' Cognizant's Chief AI Officer, Babak Hodjat, added that it will still take more than a year before we completely see the impact of modern AI technologies on the workforce. The honest answer is that the 47.9 percent AI-attributed figure is a number companies reported — it mixes genuine AI-driven restructuring with AI-as-justification for cost-cutting decisions that may have been made on other grounds. Both the real displacement and the PR framing are happening simultaneously, and untangling them precisely is not possible from public data alone. Source: Tom's Hardware, April 2026.

What Anthropic's own research says about this is important — and more measured than either the alarm or the dismissal. In March 2026, Anthropic published a study titled 'Labor market impacts of AI: A new measure and early evidence,' authored by economists Maxim Massenkoff and Peter McCrory. The study's central finding: actual AI adoption in the workplace is still only a fraction of what AI is theoretically capable of performing. The most AI-exposed occupational sectors are Computer & Mathematical, Office & Administrative Support, Business & Financial, and Sales. But across most sectors, real-world deployment of AI is far behind theoretical capability. The study found no systematic increase in unemployment for highly exposed workers since late 2022 — though it did find suggestive evidence that hiring of younger workers has slowed in exposed occupations. Source: Anthropic research, 'Labor market impacts of AI,' March 2026; Fortune, March 6, 2026; HR Dive, March 2026.

The United States IT Market: What the Data Shows for American Tech Workers

The United States technology sector is the largest and most AI-exposed in the world. It is also the sector where the impact of AI capability improvements translates most directly into workforce decisions, because US technology companies have the capital to invest in AI tools, the incentives to reduce labor costs where AI can substitute, and the global market position that makes any efficiency gain immediately significant at scale. Source: Fortune, April 2026; Anthropic Economic Index March 2026 report.

• Oracle's Q1 2026 restructuring: Oracle cut more than 10,000 employees globally in early 2026 — with a significant portion of those in India — as part of a plan to reduce expenses by hundreds of millions of dollars while simultaneously increasing investment in AI infrastructure. Oracle framed this explicitly as a shift toward AI-driven workflows. For US-based Oracle employees, the cuts were concentrated in roles that involved manual data processing, routine software maintenance, and customer support functions — precisely the tasks that AI tools are most capable of performing. Source: BusinessToday, April 2026; eWeek, April 2026.
• Bloomberg's Q1 2026 data: Bloomberg and multiple layoff trackers reported that more than 50,000 US tech employees were laid off in the first three months of 2026 alone. This is a layoff rate that, annualized, would represent the largest technology workforce contraction since 2020. The jobs most commonly affected are those at the intersection of the skills that AI tools have most rapidly improved: code generation, software testing, data analysis, technical writing, and customer support automation. Source: Bloomberg, cited by eWeek, April 2026.
• Anthropic's Economic Index on coding migration: Anthropic's own March 2026 Economic Index report found that coding tasks are 'migrating from augmentative usage in Claude.ai to more automated workflows in first-party API traffic.' In plain terms: what began as developers using AI to go faster is increasingly becoming AI running coding workflows autonomously, with human review rather than human execution. The top 10 tasks in Anthropic's API traffic represent 32 percent of all API usage — up from 28 percent just months earlier — indicating increasing concentration in high-automation, low-human-intervention workflows. Source: Anthropic Economic Index March 2026 report.
• The roles most at risk: The specific roles identified by Anthropic's economic research as most exposed to AI displacement — measured by what AI is actually doing today, not just what it theoretically could do — are concentrated in software development tasks requiring less expertise (junior and mid-level coding), office and administrative support (data entry, document processing, correspondence management), business and financial analysis (routine report generation, data synthesis, structured analysis), and sales support (lead qualification, follow-up drafting, CRM management). Workers with higher expertise and the ability to direct, review, and extend AI outputs are less exposed — and are increasingly in demand. Source: Anthropic, 'Labor market impacts of AI,' March 2026.
• The counterbalancing signal: The same Anthropic economic research found that high-tenure AI users — people who have developed habits and strategies for using Claude effectively — attempt higher-value tasks, achieve more successful responses, and are more productive than new users by a margin that compounds over time. The workers who are learning to direct AI effectively are not being displaced by it. The workers who are not adapting — who are performing the same tasks the same way without incorporating AI into their workflows — are facing real displacement risk. This is not a morally comfortable finding, but it is what the data shows. Source: Anthropic Economic Index March 2026 report.

The Indian IT Market: The Largest and Most Acute AI Disruption Story in the World

India's technology sector is the world's largest provider of software services, employing an estimated 5 million people directly and supporting tens of millions more in related industries. Its business model — built on providing skilled software engineering labor at competitive costs to US and European enterprise clients — is the business model most structurally exposed to the kind of AI capability improvement that Opus 4.7 represents. When an AI model can resolve nearly nine in ten real GitHub bugs, the cost calculus of offshore software services changes in ways that are not incremental. They are foundational. Sources: Rest of World, February 2026; CNBC, August 2025; UnboxFuture, April 2026.

The numbers for India's IT sector in 2025 and early 2026 are among the most significant labor market data points in the global AI story. Tata Consultancy Services (TCS) — India's largest private-sector employer, with a workforce that exceeded 600,000 people — cut approximately 20,000 jobs in fall 2025, explicitly citing AI-driven restructuring as a central reason. This was TCS's largest single workforce reduction in its history. The company's management framed it not as a temporary cost-cutting measure but as a permanent restructuring toward a model that requires fewer people to do the same work. The Indian tech union leader VJK Nair put it plainly: 'With artificial intelligence, the industry is getting a new challenge.' Several other major Indian outsourcers, including Infosys, Wipro, and HCL, announced their own restructuring programs in the same period, collectively affecting tens of thousands more workers. Sources: Rest of World, February 2026; CNBC, August 2025.

The WARN notice data — regulatory filings required when US-based employers conduct mass layoffs — tells an even sharper story for Indian IT companies operating in America. An analysis of regulatory filings by UnboxFuture in April 2026 found that Indian IT and BPO firms had already filed more WARN notices in the first quarter of 2026 than in all of 2025. Indian IT giants TCS, Infosys, and Wipro collectively eliminated over 5,000 US positions in recent months, with more cuts expected. The report describes this as a 'structural shift' rather than a cyclical downturn: for decades, Indian IT companies deployed thousands of engineers onsite to manage large outsourcing deals. In 2026, AI is making that model economically unviable. The outsourcing deals themselves are changing: large transformation projects that previously required hundreds of engineers are being restructured around AI-augmented workflows that require a fraction of the headcount. Source: UnboxFuture, April 2026.

India produces over 1.5 million engineering graduates annually. The IT sector has historically absorbed a significant portion of this talent. As CNBC reported in August 2025, the slowdown in IT hiring has ripple effects that extend far beyond the technology sector itself — into real estate markets in Bengaluru and Hyderabad, into the service economies that surround tech parks, and into the educational pipeline, where the signal that IT jobs are contracting has not yet fully reached the millions of students currently enrolled in engineering programs. The gap between engineering graduates entering the labor market and engineering jobs available to them is widening in a way that India has not experienced in recent memory. Source: CNBC, August 2025; Rest of World, February 2026.

Beyond India and the US: How Other Technology Markets Are Being Affected

The US and India command the most attention because they are the two largest IT labor markets and the two economies where the impact of AI on software services is most acute. But the disruption is not limited to them. The pattern of AI-driven restructuring in technology is visible across every major economy with a significant software services sector. Source: Rest of World, January 2026; eWeek, April 2026.

• Eastern Europe (Poland, Romania, Ukraine): Eastern European countries developed significant technology outsourcing sectors in the 2010s by offering skilled engineering talent at lower costs than Western Europe, positioning themselves as alternatives to Indian IT outsourcing. These markets are facing a double pressure: AI tools reduce the price advantage of lower-cost labor, and the wartime disruption of Ukraine's economy — previously one of the strongest software services sectors in the region — has compressed the talent pool. Polish and Romanian IT companies are responding by moving up the value chain toward AI-specific skills, but the transition is not instantaneous and not available to all displaced workers.
• Southeast Asia (Philippines, Vietnam): The Philippines is the world's largest business process outsourcing market, with approximately 1.5 million BPO workers. Many of the tasks these workers perform — customer service calls, data entry, document processing, basic software testing — are among the most automatable by current AI systems. The Philippine BPO sector has acknowledged that AI represents an existential challenge to its current model and has been working with government agencies on workforce transition programs, but the scale of potential displacement significantly exceeds the capacity of those programs. Vietnam's software outsourcing sector faces similar pressures, though its focus on more complex development work offers somewhat more insulation.
• United Kingdom: The UK's technology sector has seen a series of AI-attributed restructurings in 2025 and 2026, particularly in fintech, professional services technology, and media companies. The UK's strong AI policy framework — including the AI Safety Institute at DSIT — has positioned the country as a leader in AI governance but has not insulated its workers from AI-driven automation. Several major British banks explicitly cited AI automation in 2025 workforce announcements that reduced technology and operations headcount.
• Germany and continental Europe: Germany's technology sector, concentrated in industrial software, automotive AI systems, and enterprise software, is restructuring more slowly than the US and UK, partly due to stronger labor protections that make rapid workforce changes more legally complex. However, the underlying pressure is identical: AI tools are performing tasks that previously required large teams of engineers. German companies appear to be managing the transition through attrition — slowing hiring rather than conducting mass layoffs — but the net effect on employment is similar over a longer timeframe.

Are We Heading Toward Destruction? The Honest Answer Is More Complicated Than Either Side Wants.

The question deserves a direct answer, not corporate reassurance or catastrophist panic. Let's be clear about what is actually happening, what is uncertain, and what the evidence does and does not support. Source: Anthropic Economic Index research, March 2026; Fortune, April 2026; multiple cited sources.

What is demonstrably true: AI coding tools — including Claude Opus models — are performing tasks that previously required human engineers, at a level of quality that is improving faster than most experts predicted. The jobs most clearly at risk are entry-level and routine software development roles: debugging, test writing, code documentation, basic feature implementation, and repetitive data analysis. These roles are not disappearing overnight, but the pipeline that feeds them — the progression from junior engineer to senior engineer that has sustained millions of careers — is under pressure. When AI can do what a junior engineer does at a fraction of the cost, companies hire fewer junior engineers. When they hire fewer junior engineers, they have fewer senior engineers in five years. This is the 'broken ladder' problem that several economists have flagged. Sources: Anthropic Economic Index; Fortune, March 2026; Rest of World, January 2026.

What is uncertain: whether AI productivity gains will create enough new work — new products, new markets, new categories of software that could not have existed without AI tools — to absorb displaced workers. The historical record on technology transitions is genuinely mixed. The ATM did not eliminate bank teller employment; it expanded banking services enough that teller employment actually grew for decades after ATM introduction. The word processor did not eliminate secretarial work; it transformed it. But previous technology transitions happened over decades. The current AI transition is happening in years, and the breadth of its impact across different occupational categories simultaneously is without historical precedent. The honest answer to 'will it create more jobs than it destroys' is: we don't know yet, and anyone who claims certainty is overstating what current evidence supports. Sources: Anthropic Economic Index March 2026; multiple cited labor economists.

What the Anthropic research specifically found — and this is the most rigorous data available because it comes from observing actual AI use in actual workplaces — is that AI is both augmenting and replacing workers, and the proportion between these two outcomes depends heavily on skill level, adaptability, and whether workers and employers are investing in AI skills development. Workers who have learned to use AI tools effectively and direct them toward complex work are more productive and more valuable. Workers performing routine tasks that AI can now automate without human direction face real displacement risk. The transition is not destroying work. It is sorting it — dramatically rewarding the workers who learn to work alongside AI and penalizing those who don't. Whether that sorting is acceptable depends on your view of economic fairness and on whether retraining programs and policy frameworks can keep pace with the speed of change. The evidence on the second question is not encouraging. Sources: Anthropic Economic Index March 2026; Fortune, April 7, 2026.

The Jobs Claude Opus 4.7 Cannot Do — Yet

Claude Opus 4.7's benchmark scores are impressive. Its SWE-bench performance represents a real capability for routine and moderately complex software engineering. But there are specific categories of work where, as of April 2026, it cannot substitute for experienced human professionals — and understanding these boundaries is essential for calibrating both risk and opportunity.

• Novel system architecture decisions: Deciding how to structure a large, novel software system — what tradeoffs to make, which technical debt is acceptable, how to balance competing requirements that have no obvious right answer — requires judgment that AI models currently approximate but do not reliably replicate. SWE-bench tests whether a model can fix existing bugs in existing codebases. It does not test whether the model can design a system from scratch with the constraints, politics, and organizational context that real engineering decisions involve. Senior architects and technical leads who make these judgment calls are not yet at risk.
• Customer and stakeholder relationship management: Software development happens in the context of human relationships — clients with changing requirements, internal stakeholders with competing priorities, users whose behavior and needs cannot be fully specified in advance. The work of translating human needs into technical requirements, managing expectations, and navigating organizational dynamics is not automatable by any current AI system. Technical project managers, product managers, and engineering leads who function at this interface remain essential.
• Frontier research: Claude Opus 4.7 scores 94.2% on GPQA Diamond — a benchmark of graduate-level scientific reasoning. This is impressive. It is not the same as conducting original scientific research. The judgment required to identify which questions are worth asking, which experimental designs are most likely to produce useful knowledge, and how to interpret unexpected results in the context of a field's open problems remains beyond current AI capability.
• Regulatory, legal, and compliance-heavy contexts: AI systems can draft contracts, summarize regulations, and identify compliance issues — but they cannot bear the professional liability that attorneys, compliance officers, and regulated professionals carry. In domains where an error has legal consequences and where a licensed professional must certify work, the human role is protected not just by capability but by regulatory structure.
• Hardware, physical systems, and manufacturing: Software that controls physical systems — from medical devices to manufacturing equipment to vehicles — must be validated in physical environments. The testing, debugging, and certification processes for safety-critical embedded software involve physical constraints that AI cannot navigate remotely. The engineers who work at the intersection of software and hardware systems are among the least exposed AI displacement categories in the technology sector.

What the Future Actually Looks Like: Three Scenarios for 2027 and Beyond

Forecasting AI capability trajectories is genuinely difficult, and anyone presenting a single certain future is telling you what they wish were true rather than what evidence supports. Here are three scenarios that reflect the range of credible expert opinion, with the evidence that supports each.

Scenario 1 — Augmentation Equilibrium (Moderate Probability)

In this scenario, AI capability continues to improve but the rate of real-world displacement is moderated by several factors: the adaptation of workers who successfully incorporate AI tools into their practice, the creation of new categories of software and digital products that AI tools make possible, the regulatory environment that develops around AI use in high-stakes domains, and the organizational friction that slows enterprise adoption. The net labor market effect is negative for specific categories of entry-level knowledge work but not catastrophic in aggregate. Wage compression occurs in highly AI-exposed roles. New premium roles emerge for people who can effectively direct AI systems on complex tasks. India's IT sector contracts but pivots partially toward higher-value AI-integrated services. This is the scenario that Anthropic's economic research, taken at face value, most closely supports as of early 2026.

Scenario 2 — Accelerating Displacement (Lower But Non-Trivial Probability)

In this scenario, the pace of AI capability improvement exceeds the pace at which workers and institutions can adapt. Models like Opus 4.7 are replaced by Opus 4.8, 4.9, and Claude 5 at approximately two-month intervals, each one covering more of the capability space that currently requires human judgment. The 'broken ladder' effect compounds: fewer junior engineers means fewer senior engineers in five years, which means fewer architects in ten, which means the institutional knowledge required to oversee complex AI-built systems becomes scarce at precisely the moment when that oversight is most needed. Enterprise AI adoption, driven by cost pressure and competitive dynamics, accelerates faster than retraining infrastructure can handle. Unemployment in highly exposed sectors rises materially — not to the catastrophic levels sometimes predicted, but enough to represent a genuine social and economic crisis concentrated in specific demographics and geographies. This is the scenario that Alex Stamos's six-month warning on cybersecurity (in the Project Glasswing context) implicitly suggests is possible: that capability thresholds can be crossed faster than institutions respond.

Scenario 3 — Structural Transformation With New Jobs (Optimistic, Historically Supported, But Uncertain)

In this scenario, AI tools generate enough new economic activity to absorb displaced workers, but not in the same roles, not in the same companies, and not necessarily in the same geographies. New categories of work emerge that today's frameworks do not capture — AI trainers, alignment specialists, model evaluators, AI-native product categories that create demand for new skills. The analogy is not the ATM but the smartphone: the smartphone eliminated many mobile manufacturing and landline telephone jobs while creating enormous new industries — mobile apps, the app economy, location-based services, social media — that employed far more people. Whether AI has the same expansionary effect on net employment is the central empirical question of the decade, and the answer is not yet knowable from Q1 2026 data. The most honest framing is: this is the scenario that history suggests is possible, and the evidence from previous technology transitions offers some support, but the speed and breadth of current AI adoption is genuinely without precedent.

5 Career Moves That Actually Matter Right Now — Based on the Data, Not on Reassurance

Most career advice written in response to AI scares is useless because it is too generic and arrives too late. What follows is derived specifically from what Anthropic's economic research, real labor market data, and the specific capability profile of Opus 4.7 actually show — not from the reassurance that gets recycled every time AI news spikes. Five moves. Do them in the next 30 days, not the next year.

Move 1: Learn to Direct AI, Not Just Use It

Anthropic's Economic Index found that high-tenure AI users attempt higher-value tasks and succeed more often than new users — by a compounding margin. The skill being rewarded is not the ability to use an AI tool, which is accessible to everyone. It is the ability to direct AI toward complex work: decomposing hard problems into sequences an AI can execute, evaluating AI outputs with enough domain knowledge to catch errors, and escalating to human judgment when AI outputs cannot be trusted. This is the difference between a user and a director. The labor market is increasingly rewarding directors. Source: Anthropic Economic Index March 2026.

Move 2: Move Into the Interface Between AI and Physical Reality

The jobs least exposed to AI displacement, according to Anthropic's observed exposure data, are those involving physical presence, physical world interaction, and regulatory liability in high-stakes domains. Embedded systems engineering, medical device software, hardware-software integration, and safety-critical systems validation all involve physical constraints that remote AI operation cannot satisfy. If your career currently sits entirely in software that operates on computers to produce other software, consider whether there are pathways into domains where software meets physical systems, regulatory frameworks, or professional liability — domains where AI augments rather than replaces.

Move 3: Build Expertise in What AI Gets Wrong, Not Just What It Gets Right

Opus 4.7 resolves 87.6% of SWE-bench bugs. The 12.4% it does not resolve — the bugs that require judgment about system architecture, domain-specific constraints, edge cases that the training distribution did not cover, or problems that require understanding the human context behind the code — is where human expertise remains essential. Building deep expertise in the specific failure modes of AI coding systems makes you increasingly valuable as AI handles routine work, because you become the person who can evaluate, correct, and extend what AI produces. The engineer who understands where AI is likely to go wrong is more valuable than the engineer who simply generates AI outputs.

Move 4: For Indian IT Professionals — The Pivot Is Now, Not Later

The structural shift in Indian IT is real and is accelerating. The business model of large-scale labor arbitrage in routine software development is being compressed by AI faster than most industry timelines predicted a year ago. The pivot that TCS, Infosys, and Wipro are making — toward higher-value AI-integrated services, toward AI implementation and governance roles, toward domain-specific expertise in industries like finance, healthcare, and legal — is the correct pivot. But the companies making that pivot cannot absorb all the workers displaced from the previous model. Indian IT professionals currently in routine software maintenance, data entry, BPO, or basic development roles face the most acute near-term displacement risk. The window for skill transition — to AI implementation, to domain expertise, to the specialized roles that AI augments rather than replaces — is open right now. It will narrow. Sources: Rest of World, February 2026; UnboxFuture, April 2026.

Move 5: Follow the Anthropic Economic Research, Not the Headlines

The most accurate ongoing picture of AI's real impact on the labor market — not the theoretical potential, not the catastrophist projections, not the corporate reassurance — is Anthropic's Economic Index research. It is published quarterly, based on observed usage of Claude in real enterprise environments, and it measures what AI is actually doing in the economy as opposed to what models are capable of in laboratory conditions. The gap between capability and actual deployment is consistently larger than AI coverage suggests, and it is changing month over month. Following this research does not guarantee perfect career decisions, but it provides better signal than headline coverage of individual model releases, including this one. Source: Anthropic Economic Index, quarterly publication.

Pricing and Access: Who Can Use Opus 4.7 and What It Costs

Claude Opus 4.7 is available as of April 16, 2026, through all of Anthropic's primary channels. For individuals, it is accessible through Claude Pro ($20/month), Claude Max ($100/month for higher usage limits), Claude Team, and Claude Enterprise tiers. For developers and businesses building on the API, the pricing is unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens. Batch processing is available at 50% discount ($2.50/$12.50 per million tokens). Prompt caching (for frequently reused context) offers up to 90% cost savings. The 1 million token context window carries no long-context premium — a million-token request costs the same per-token rate as a 9,000-token request. Source: APIdog, April 16, 2026; BenchLM.ai, April 16, 2026; Anthropic pricing page.

Opus 4.7 is also available through GitHub Copilot (replacing Opus 4.5 and 4.6 in the model picker for Copilot Pro+, Business, and Enterprise over the coming weeks), through Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. The breadth of distribution — including the GitHub Copilot integration, which reaches millions of software developers through their existing development environments — means that the capability increase in Opus 4.7 will be in the hands of a very large number of software engineers within days of release. Source: GitHub Changelog, April 16, 2026; Anthropic official announcement.

Frequently Asked Questions