Enterprise organizations are shifting agentic AI from experimental pilots to production environments to automate complex workflows. While companies like Automation Anywhere and Yaskawa Electric report significant operational wins, a systemic reality gap
persists, with only 11% of organizations actively using these systems in production according to a 2025 Deloitte study.
The transition is no longer about whether a model can generate a clever response, but whether it can execute a business process without constant human intervention. For some, the scale is already massive. Automation Anywhere’s Autonomous Service Desk has fulfilled more than one billion IT service requests, auto-resolving an average of over 80% of them.
But for the broader market, the move to production is hitting a wall. This is mirrored by Gartner’s finding that at least 50% of generative AI projects are abandoned after the proof-of-concept stage due to escalating costs, poor data quality, or inadequate risk controls.
The Silicon-Based Workforce and Legacy Friction
Industry leaders are beginning to categorize AI agents not as tools, but as a silicon-based workforce
. This shift requires a fundamental redesign of how work is structured. For example, Toyota is using agentic tools for real-time visibility of vehicle arrivals at dealerships and plans to let agents resolve supply chain issues by bypassing human interaction with complex mainframe systems.
However, this digital labor force is often crippled by the very systems it is meant to optimize. Gartner forecasts that more than 40% of agentic AI projects will fail by 2027 because of legacy system incompatibility. Many firms make the mistake of layering agents onto old, human-centric processes—a failure that can lead to workslop
, where poorly designed applications actually increase the operational burden.
To combat this, some organizations are restructuring their entire corporate hierarchy.
Industrial Execution: Yaskawa, KUKA, and Automation Anywhere
While software firms tackle the back office, robotics manufacturers are pushing agentic logic onto the factory floor. Yaskawa Electric has developed a robot that integrates with Google DeepMind’s Gemini to assess site conditions and write its own work procedures. This allows the system to reason and adapt without a programmer needing to re-code for every single variable.
Meanwhile, Automation Anywhere is seeing the financial impact of this shift; the number of enterprises with more than $1 million in Annual Recurring Revenue (ARR) grew 25% year over year.
| Entity | Production Application | Verified Outcome/Goal |
|---|---|---|
| Automation Anywhere | Autonomous Service Desk | 80%+ auto-resolution rate across 1B+ requests |
| Yaskawa Electric | Gemini-integrated Robots | Self-executing work procedures based on site conditions |
| Toyota | Supply Chain Agents | Direct resolution of issues via mainframe bypass |
| Mapfre | Claims Management | Automation of routine administrative tasks |
Engineering the Path to Scale: FDE and AI Hubs
The gap between a pilot and a production-ready system is often an engineering problem rather than a model problem. To bridge this, ServiceNow and Accenture have launched a Forward Deployed Engineering (FDE) program. Instead of providing instructions from afar, their teams embed directly into customer environments to build workflows natively on the ServiceNow AI Platform.
Other vendors are focusing on the hardware-software stack to reduce infrastructure friction.
ASUS has introduced the AI Hub, a turnkey solution that combines language models and prompt libraries with NVIDIA-certified server infrastructure. By supporting on-premises deployment, the hub allows sensitive data to remain local, addressing the governance and security fears that often stall cloud-based AI projects.
These allow specialized agents to collaborate across entire workflows rather than acting as isolated bots.
The Governance Bottleneck
As autonomy increases, the question of accountability becomes paramount. Infobip’s CTO, Izabel Jelenić, argues that while agents can act with increasing autonomy, business accountability must remain with the organization. Every agent requires a defined purpose, a business owner, and strict boundaries regarding which systems it can access and which permissions it holds.
Financial risk is also evolving. Without these guardrails, the efficiency gains of a silicon workforce
could be erased by unpredictable operational expenses.
The stakes are high. Global AI spending is projected to exceed $2 trillion in 2026. Yet, as long as only a fraction of organizations have deployable solutions, the industry remains in a state of tension between massive capital investment and actual operational utility.
The remaining uncertainty lies in the “human-agent” handoff. While companies like Mapfre keep humans in the loop for sensitive interactions, the industry has yet to standardize how an agent recognizes its own limits and escalates a problem to a human without creating a new layer of operational friction.
Related reading
