Amazon hires Adept leadership for AGI push
Amazon hired Adept co-founder David Luan and several other co-founders to lead its AGI Autonomy segment. This move follows Adept’s struggle to launch successful commercial products despite raising over $400 million in venture capital since its 2022 founding. Amazon pays Adept a licensing fee to use its AI agent technology to automate business tasks. The Federal Trade Commission is requesting details from Amazon regarding its deal to hire top executives and researchers from Adept, an inquiry that mirrors the regulator’s scrutiny of Microsoft’s agreement to license technology from Inflection AI. Zach Brock, the former head of engineering, now leads Adept as CEO. Luan will report to Rohit Prasad, the head of the AGI team, who reports to Amazon CEO Andy Jassy. I view this pivot as a way for Adept to focus on its agent vision while Amazon builds its own large language models. Amazon also holds a $4 billion investment in the AI startup Anthropic. About 20 employees remain at the startup.
Agentic AI replaces failing RPA workflows
The market is moving away from Robotic Process Automation (RPA) toward agentic AI workflows. RPA relies on rigid scripts that mimic human actions like copying data or clicking buttons. Ernst and Young reports a 30 to 50 percent failure rate for RPA bots when software updates change the user interface. Agentic AI differs because it plans, acts, and verifies tasks with autonomy.
| Entity/Use Case | Metric | Result |
|---|---|---|
| Equinix (E-Bot) | Resolution Time | 5 hours to 30 seconds |
| Dutch Insurer (Beam) | Autonomous Claims | 91% |
| H&M Assistant | Customer Queries | 70% human-free |
| Bank of America (Erica) | Support Calls | 17% reduction |
You should recognize that these numbers prove the efficiency of goal-driven agents. Equinix uses its E-Bot to triage IT tickets with 96 percent accuracy, which cut resolution times from five hours to 30 seconds. H&M reports a 25 percent increase in conversion rates using its agentic shopping assistant that handles 70 percent of customer queries without humans. A Dutch insurer with Beam’s system processes 91 percent of claims autonomously and sees a 46 percent increase in processing speed. Bank of America’s Erica assistant manages over 1 billion user interactions and reduces human support calls by 17 percent. The current failure rate of RPA bots makes them an unreliable choice for modern enterprises.
Labor shortages drive the need for these tools in sectors like manufacturing. Honeywell CEO Vimal Kapur notes that customers in hospitals and data centers struggle to find qualified technicians. Many businesses implement agentic workflows for tasks like automated customer onboarding, dynamic lead qualification, and predictive inventory control.
Measuring the move to autonomous enterprises
Enterprises struggle to move past the early levels of autonomy. Many manufacturers, specifically 80 percent of U.S. facilities, operate with zero automation today. Gartner predicts that more than 40 percent of agentic AI projects will fail by the end of 2027 due to rising costs or weak risk controls. Organizations attempt to move from Level 1, which uses assistive co-pilots, toward Level 3, where multiple agents coordinate across functions.
The ability of an agent to handle a desktop environment determines its real utility. Coasty scores 82 percent on the OSWorld benchmark, which tests 369 real desktop tasks involving file management and web browsing. Anthropic’s Claude 4.5 Sonnet scores 61.4 percent on that same benchmark. This 20 percent gap separates tools that work from those that merely demonstrate research. Will companies successfully manage the risk of autonomous decisions in highly regulated sectors?
