The artificial intelligence industry has entered a new phase. For years, the dominant assumption was that better models alone would unlock a productivity revolution. The logic appeared straightforward: if AI systems could write, code, summarize, analyze and automate cognitive tasks faster than humans, then economic output would inevitably surge. Yet despite the explosive adoption of generative AI tools across industries, the long-promised productivity boom has largely failed to materialize.
This gap between technological capability and measurable economic impact has become the defining puzzle of the AI era.
Now, the world’s leading AI companies are openly acknowledging the problem. Both OpenAI and Anthropic are moving beyond simply selling access to large language models. Instead, they are building private equity-backed consulting operations designed to restructure how companies actually work. Their objective is no longer merely to provide AI tools but to redesign organizational systems so those tools can generate measurable output gains.
The shift is significant because it reflects a growing realization across the industry: the true bottleneck in AI adoption is not model intelligence. It is organizational architecture.
Reports indicate that Anthropic is raising approximately $1.5 billion with support from firms including Blackstone, Hellman & Friedman and Goldman Sachs to build enterprise deployment capabilities. OpenAI, meanwhile, has reportedly secured around $4 billion from investors led by TPG, Brookfield and Bain Capital at a valuation near $10 billion for a similar initiative.
Together, these efforts represent more than a fundraising trend. They reveal an industry-wide recognition that the “AI productivity paradox” cannot be solved by scaling models alone.
The paradox is striking. AI usage in the workplace has expanded dramatically in only a few years. Surveys show that a growing share of workers regularly use generative AI systems for writing, coding, research and communication. Yet overall productivity growth in advanced economies has remained sluggish.
This contradiction resembles the famous information technology productivity paradox of the 1980s and 1990s. During that period, computers spread rapidly through offices and factories, but economic productivity statistics barely moved for years. Only after companies fundamentally reorganized themselves around digital systems did the gains begin to appear.
The same pattern may now be repeating with AI.
The central issue is that individual efficiency does not automatically translate into organizational productivity. A worker may complete tasks faster using AI, but unless the broader workflow changes, the company itself may not produce more output.
A customer service representative assisted by AI might answer emails more quickly. A lawyer might draft contracts in less time. A programmer might generate code faster. Yet if management structures, approval chains, communication systems and organizational incentives remain unchanged, the firm as a whole may continue operating at the same speed.
This is where the concept of “organizational plasticity” becomes important. Firms capable of rapidly redesigning workflows, redistributing responsibilities and integrating modular processes are more likely to convert AI-driven task efficiencies into actual economic gains. Firms that lack this flexibility may simply experience marginal improvements in employee convenience rather than transformational productivity growth.
Research increasingly supports this view. Studies on AI adoption often demonstrate strong improvements at the task level but weaker effects at the system level. Workers save time, but companies do not necessarily produce more.
One large experiment involving Microsoft 365 Copilot illustrated the problem clearly. Employees using AI tools spent less time on email and documentation, but meeting loads remained constant and overall work output changed little. Many workers simply reclaimed personal time instead of increasing production. The machine became more efficient internally, but the organization itself did not move faster.
This distinction is now reshaping the business strategies of frontier AI firms.
The emergence of AI consulting arms is effectively an admission that AI adoption is not a plug-and-play process. Selling API access or subscriptions is insufficient if organizations cannot restructure themselves around AI capabilities.
As a result, leading AI companies are beginning to resemble enterprise transformation firms as much as software providers.
Their new business model involves embedding engineers directly inside client organizations, redesigning workflows, integrating AI systems into sector-specific operations and replicating successful deployment frameworks across multiple firms. The objective is to raise the “plasticity” of companies from within.
This approach is expensive, labor-intensive and slow. But it may also be necessary.
The economic value of AI increasingly appears to reside not in the models themselves but in the operational systems surrounding them. The frontier labs have discovered that intelligence alone does not create productivity. Coordination does.
This realization has profound implications for the future structure of the technology industry. The next dominant AI companies may not simply be those with the largest models or most advanced chips. They may be the firms best able to redesign institutions, workflows and labor systems.
While American firms focus on transforming existing corporations from the inside, China appears to be experimenting with a more radical idea: bypassing the traditional firm altogether.
The rise of the open-source AI agent OpenClaw in China reflects this alternative model. Once the software escaped niche developer communities and reached mainstream users, Chinese technology companies rapidly integrated it into broader digital ecosystems.
Companies such as ByteDance, Tencent and Alibaba Group moved quickly to build AI agent infrastructure into cloud platforms, messaging apps and enterprise tools.
Because the underlying model was open source, competitive advantage shifted away from the model itself and toward distribution, integration and workflow orchestration. The winning strategy became embedding AI directly into daily digital life.
At the same time, Chinese local governments began subsidizing AI adoption not primarily for corporations, but for individuals and micro-enterprises. Cities and development zones introduced compute vouchers, AI entrepreneurship incentives and support frameworks for one-person businesses powered by AI agents.
The logic behind this strategy is straightforward. China already possesses extensive manufacturing networks, logistics systems, payment infrastructure and digital marketplaces. AI agents can allow individuals to plug directly into these systems without needing large organizational structures.
A single entrepreneur with AI assistance can theoretically coordinate suppliers, manage customer communication, automate marketing and operate across digital commerce ecosystems with minimal overhead.
In this vision, AI reduces the minimum viable size of a productive organization.
Instead of transforming existing firms, AI enables entirely new forms of decentralized production.
The United States and China are therefore converging on two distinct models for solving the AI productivity puzzle.
The American model assumes that large incumbent firms remain the core engines of economic activity. The challenge is making those firms more adaptable. AI consultants, embedded engineers and workflow redesign specialists become essential intermediaries in this process.
The Chinese model assumes that the firm itself may be the problem. Rather than increasing organizational flexibility inside corporations, China is experimenting with lowering the organizational threshold required for production altogether.
One path emphasizes institutional transformation.
The other emphasizes institutional fragmentation.
Both strategies carry significant risks.
The American approach may struggle with scale. Organizational change is notoriously difficult, especially inside large bureaucratic firms. Embedding engineers into thousands of companies could prove extraordinarily expensive and operationally inefficient.
The Chinese approach faces a different danger: a flood of low-quality automation and speculative micro-businesses with weak economic durability. Many AI-enabled solo ventures may fail to generate sustainable value.
Yet despite their differences, both strategies are attempting to solve the same underlying challenge: converting task-level efficiency gains into system-level economic output.
The global AI race is often framed around semiconductors, data centers and model benchmarks. Those elements remain important. But they may no longer be the decisive factor.
The harder challenge lies in designing the social and organizational systems through which AI is deployed.
History suggests that transformative technologies do not reshape economies immediately. Electricity existed for decades before factories reorganized themselves around electrified production lines. Computers spread long before companies redesigned workflows around digital networks.
AI may follow the same trajectory.
The next breakthrough may not come from a more powerful chatbot or a larger model. It may emerge from a better organizational structure, a new labor model or an entirely different conception of the firm itself.
The real AI revolution is no longer about teaching machines to think. It is about redesigning the systems through which humans work.
Abul Quashem Joarder, a contributor to Blitz is geopolitical and military expert.