Highlight of the Week: AI Agents, from Experimentation to Implementation
This week also saw a series of remarkable developments in the field of artificial intelligence (AI). In particular, the period from October 12 to 18, 2025, marked a historic turning point in what could be called the “first year of AI agents,” when “AI agents,” which had previously been in an experimental stage, clearly moved to the next stage of enterprise implementation.
From massive infrastructure investments to tangible business applications to new technologies that are revolutionizing the manufacturing workplace, a series of innovations and business developments announced over the past week are bringing unprecedented transformational potential to the manufacturing industry. In this article, we organize the major AI-related news announcements from around the world during this period and take an in-depth look at their practical applicability, particularly in the manufacturing industry.
1. competition for “gigawatt-class” investment in AI infrastructure
OpenAI and Broadcom to Achieve Self-Improvement Cycle for AI Chip Design
On October 13, OpenAI and Broadcom announced a strategic alliance to deploy AI accelerators at a staggering 10 gigawatts. At the heart of this partnership is OpenAI’s groundbreaking approach where the AI itself is responsible for designing the next generation of chips; a “self-improvement cycle” where AI systems design the hardware to support themselves is becoming a reality.
This 10 gigawatts equates to $500-600 billion worth of infrastructure value, based on NVIDIA CEO estimates.
Implications for Manufacturing:
This move will accelerate the diversification of AI infrastructure procurement strategies in the manufacturing industry, moving away from a one-party NVIDIA model and diversifying across multiple vendors, including AMD and Broadcom, allowing for cost optimization and flexibility in infrastructure selection. Manufacturing executives should develop mid- to long-term data center strategies in anticipation of these large-scale infrastructure deployments scheduled for the second half of 2026.
Oracle Zettascale10: Largest AI Supercomputer Ever in the Cloud
On October 14, Oracle announced the Zettascale10, which connects up to 800,000 NVIDIA GPUs and boasts a peak performance of 16 zetaFLOPS. It is the largest cloud-based AI supercomputer ever deployed across multiple data centers.
Implications for Manufacturing:
Large-scale simulations, materials science calculations for new materials development, production process optimization, and other computation-intensive tasks that have been difficult to perform with conventional in-house supercomputers can now be easily performed in the cloud. The shift to a pay-as-you-go model that enables the use of computing resources as and when needed will accelerate.
2. enterprise AI agent, evolving into a job executor
Google Gemini Enterprise: A Strategic Development to Engage the Competition
On October 9, Google announced Gemini Enterprise, a competitor to Microsoft 365 Copilot at a competitive price of $21-$30 per month. Notably, it allows for integration with existing Microsoft 365, a strategic step that allows for the adoption of Google’s AI while maintaining the existing ecosystem.
Virgin Voyages, an early implementation case study, has already deployed more than 50 specialized AI agents, signaling a future in which multiple specialized agents, rather than a single assistant, work together to complete complex tasks.
Microsoft Agent Mode: To autonomous execution of workflow
Microsoft’s “Agent Mode,” which was rolled out in late September and October, gives AI the ability to autonomously perform multi-step tasks based on user instructions within applications such as Excel, Word, and PowerPoint. to “execute a complete workflow.
Practical Applications in Manufacturing:
AI agents will revolutionize administrative and field operations in the manufacturing industry.
- Supply chain optimization: Automatically adjusts and executes everything from inventory management to ordering and logistics arrangements in response to market demand fluctuations.
- Dynamic optimization of production planning: Production schedules are automatically created and updated in real time, taking into account multiple constraints (manpower, equipment utilization rates, material delivery dates).
- Automation of quality control reports: Analyzes inspection data in multiple formats and performs autonomous execution from report generation to proposal of corrective actions and assignment of tasks to personnel.
3. dramatic improvement in development productivity and the rise of multiple models
IBM Project Bob: 45% Improvement in Development Efficiency
On October 7, IBM and Anthropic, along with a strategic alliance, announced that Project Bob, an AI integrated development environment for 6,000 in-house developers, has achieved a 45% productivity increase.
The key to Project Bob is “multi-model orchestration,” where AI automatically uses Anthropic Claude, Mistral, Meta Llama, and IBM’s own models according to task requirements (accuracy, cost, latency) to derive optimal results. The AI automatically uses Mistral, Meta Llama, and IBM’s proprietary models according to task requirements (accuracy, cost, and latency) to derive optimal results. The fact that 95% of users use it not for code generation, but for completing complex tasks, shows that AI is evolving from a mere coding aid to a true collaborator.
Implications for Manufacturing:
- Acceleration of customized development of MES (Manufacturing Execution System) and ERP.
- Improved efficiency of PLC (Programmable Logic Controller) programming.
- Support for modernization of legacy systems and automatic detection of security vulnerabilities.
4. advanced technologies to innovate manufacturing sites: generative AI and nondestructive testing
MIT SpectroGen: The Birth of a “Virtual Spectrometer” Using Generative AI
On October 14, the MIT research team presented SpectroGen, a generative AI tool. It is a “virtual spectrograph” that can virtually generate expensive X-ray diffraction data from inexpensive infrared scan data with 99% accuracy. It replaces multiple expensive measurement devices and a material quality verification process that used to take days with a single inexpensive device and AI, reducing measurement time to less than a minute.
Manufacturing applications:
It enables faster and lower cost wafer and material quality inspections in semiconductor and battery manufacturing, eliminating bottlenecks in shortening development cycles.
Purdue University RAPTOR: Improving the Accuracy of Semiconductor Defect Detection
A research team from Purdue University has presented the RAPTOR system, which combines high-resolution X-ray imaging and machine learning to achieve 97.6% accuracy in detecting minute defects inside chips. Furthermore, the system can be applied to detect counterfeit chips in the supply chain.
Manufacturing applications:
It contributes to dramatic improvements in quality assurance in the automotive and aerospace industries, where high reliability is required, and to preventing the influx of counterfeit parts in the supply chain.
Bridging the OT-IT Gap and AI Agents
Forbes magazine published an in-depth analysis on October 17 on the integration of operational technology (OT) and IT-based systems on the manufacturing floor (closing the OT-IT gap), which is essential now that AI agents are beginning full-scale operations on the factory floor.
The “data-as-is strategy” presented in the article is to take existing PLC data and equipment logs directly into AI without major system modifications and let AI handle the integration and normalization. This provides a roadmap for gradual modernization for manufacturing companies with legacy systems.
5. practical application scenario in the manufacturing industry (after the use of AI agents)
Trends over the past week provide concrete indications of the future of work in the manufacturing industry.
| Scenarios | Conventional issues | Changes after utilizing AI agents (example) |
| Advancement of predictive maintenance | Responding to failures as they occur. Manual parts ordering and planning reorganization. | Virtual detection of material degradation with SpectroGen. Multi-model AI detects signs of abnormality at an early stage, and Agent Mode autonomously executes parts ordering, maintenance schedule adjustment, and production plan reorganization. |
| Fully automated quality control | Sampling inspections and visual checks by humans. Risk of outflow of defective products. | Total non-destructive inspection to internal defects with RAPTOR. Automatic classification and cause analysis of defective products and automatic adjustment of production parameters to reduce the defect rate. |
| Global Integrated Management | Data at each site is siloed. Language and time difference barriers make global optimization difficult. | Gemini Enterprise integrates data from all locations. Automatically monitor KPIs at each location, extract best practices, and dynamically reallocate production capacity across locations. |
6. practical steps for implementation and risk management
Companies that begin preparing now for the second half of 2026, when large-scale infrastructure deployment will be in full swing, will be the leaders of the next decade.
Implementation Steps (Recommended Roadmap)
- Status analysis (1-3 months): Identify use cases with the highest ROI (predictive maintenance, quality control, etc.); form a cross-functional pilot team including IT, OT, and field staff.
- Small-scale pilot (3-6 months): limited to a single line, technical validation of OT-IT integration. Evaluate platform based on “multi-vendor strategy”.
- Phased rollout (6-18 months): successful pilots are horizontally deployed to other departments. Transition from a single-agent to a collaborative multi-agent system.
Challenges and Countermeasures
Successful implementation of AI agents depends not only on technology but also on organizational change.
- Organizational challenge: The employment insecurity of field workers is addressed by investing in “human resource development” to help them acquire new skill sets as AI supervisors.
- Technical issues: Compatibility with legacy systems and security risks are addressed through “phased investment” and “transparency (visualization of AI decision-making process).
Conclusion: Toward a Future of AI and Human Collaboration
The week of October 12-18, 2025 was an important milestone in the evolution of AI from a mere tool to an “autonomous workflow executor” that began to penetrate deep into the enterprise.
OpenAI and Broadcom’s historic partnership is gaining ground on AI infrastructure, Google and Microsoft’s agents are making management tasks autonomous, and new technologies from MIT and Purdue University are eliminating bottlenecks in the manufacturing workplace — all these developments mean that manufacturing DX has entered a new stage.
The future of AI and human collaboration is no longer a distant dream.
Source List
Main News Source
- OpenAI and Broadcom Co-Development Partnership
OpenAI Official Blog (October 13, 2025)
https://openai.com/index/openai-and-broadcom-announce- strategic-collaboration/ - Oracle Zettascale10 Announcement
Oracle News (October 14, 2025)
https://www.oracle.com/news/announcement/ai-world-oracle-unveils-next- generation-oci-zettascale10-cluster-for-ai-2025-10-14/ - Google Gemini Enterprise Launch
Google Cloud Blog (October 9, 2025)
TechCrunch (October 9, 2025)
https://cloud.google.com/blog/products/ai- machine-learning/introducing-gemini-enterprise - IBM Project Bob and Anthropic Partnership
IBM Newsroom (October 7, 2025)
VentureBeat (October 7, 2025)
https://newsroom.ibm.com/2025-10-07-2025- ibm-and-anthropic-partner-to-advance-enterprise-software-development-with-proven-security-and-governance - Microsoft Agent Mode Introduction
Microsoft 365 Blog (September 29, 2025)
https://www.microsoft.com/en-us/microsoft-365/blog/2025/09/29/vibe -working-introducing-agent-mode-and-office-agent-in-microsoft-365-copilot/ - AI Intelligence Briefing – October 16, 2025
LinkedIn – Azumo LLC (October 16, 2025)
https://www.linkedin.com/pulse/ai-intelligence-briefing- october-16-2025-azumo-llc-jffuf - China Cambricon Earnings Announcement
Bloomberg (October 17, 2025)
https://www.bloomberg.co.jp/news/articles/2025-10-17/T49YEHGOT0JL00 - MIT SpectroGen announcement
MIT News (October 14, 2025)
https://news.mit.edu/2025/checking-quality-materials-just-got-easier-new-ai-tool-1014 - Purdue University RAPTOR System
Purdue University Newsroom (October 6, 2025)
https://www.purdue.edu/newsroom/2025/Q4/cutting-edge-imaging-ai -research-seeks-out-minus-cule-defects-in-chips - Enterprise AI Agents on the Factory Floor
Forbes – Moor Insights (October 17, 2025)
https://www.forbes.com/sites/moorinsights/2025/10/17/enterprise -ai-agents-clock-in-on-the-factory-floor/
Related Articles and Analysis
- AI Agents Just Went Enterprise
Medium – Micheal Lanham (October 2025)
https://medium.com/@Micheal-Lanham/ai-agents-just-went-enterprise- october-2025-changed-everything-dc67abb7665a - IBM AI Agents on Oracle Fusion Applications
IBM Newsroom (October 16, 2025)
https://newsroom.ibm.com/2025-10-16-ibm-announces-new-ai-agents-on -oracle-fusion-applications-ai-agent-marketplace - AI Revolutionizes Quality Control
Market Minute (October 14, 2025) - Visual AI in Manufacturing: 2025 Landscape
Voxel51 Blog (July 16, 2025)
https://voxel51.com/blog/visual-ai-in-manufacturing-2025-landscape - Japanese Manufacturing AI Applications
Nikkei Crosstech (October 16, 2025)
