Agentic artificial intelligence - AI systems that work independently to complete complex tasks - made significant strides this week with multiple major releases and deployments across the technology industry.

IBM's CUGA Agent and Enterprise Automation

International Business Machines (IBM), a major computer company, released CUGA (Configurable Generalist Agent), an open-source tool that other companies can use and modify. CUGA is designed to automate complicated business workflows that normally require human workers to complete multiple steps. The system works by orchestrating different tasks, connecting to various APIs (tools that let software communicate), and even generating its own code. In tests, CUGA achieved a 61.7% success rate on a benchmark called WebArena and 48.2% on AppWorld. While these numbers might seem low, technology experts note that these success rates could be acceptable for specific business scenarios.

However, the tech industry is showing skepticism about agents. Analysts warn that many enterprise agent projects may be canceled because companies aren't seeing enough business value from them. CUGA uses several techniques to work effectively: it detects the user's intent, maintains a task ledger to track work, delegates responsibilities to specialized agents, and follows company policies.

Google's Research Agent and API Integration

Google, the world's largest search company, released a completely reimagined Gemini Deep Research agent based on their latest Gemini 3 Pro model. Previously, this agent could only generate research reports, but now it does much more. The biggest news is that Google opened up the agent's research capabilities through a new tool called the Interactions API, which lets developers build research abilities directly into their own apps. This is important because it supports what experts call an "agentic era" - a time period where AI systems will do much more independent information seeking and decision-making.

Google says their customers use the Deep Research tool for due diligence (checking facts before making decisions) and scientific research. The company claims the new agent has better factuality and reduced hallucinations, which are when AI systems make up false information. Google even created a new test called DeepSearchQA benchmark to measure how well their agent performs.

Productivity Tools Embrace AI Agents

Notion, a tool that millions of workers use to organize their work, is expanding far beyond its original purpose. The company is positioning itself as an "everything app" for office work, potentially challenging larger companies that make suites of business tools. Notion is adding AI features including enterprise search capabilities, a knowledge work agent that can complete extended tasks, and an AI note taker. This expansion shows how AI agents are moving into everyday tools that people use for their jobs.

The Return on Investment Challenge

Despite all this innovation and investment, companies are struggling to see real benefits from their AI spending. A Reuters investigation found that many businesses are not seeing meaningful return on investment from generative AI, despite widespread adoption since ChatGPT launched in 2022. Executive surveys revealed concerning statistics: Forrester found that only 15% of companies saw profit margins improve because of AI, while BCG found just 5% experienced widespread value.

Companies reported multiple problems with their AI systems. The systems show "sycophancy," meaning they tell users what they think people want to hear rather than being honest. Performance is inconsistent - the same system might work well one day and fail the next day. Many agents fail at seemingly simple tasks that humans find easy. Handling long documents remains a major challenge.

Because of these issues, some companies are going back to using more human workers for complicated customer interactions. Artificial intelligence vendors are responding by offering more hands-on help with deployment and creating business-focused offerings.

Careful Implementation of Agents

Experts emphasize that agentic automation should be treated as assistive technology - tools that help humans, not replace them - with strong quality checks and guidelines. This is especially important for tasks that could create customer-facing mistakes or regulatory compliance problems. The message is clear: agents are improving but remain unpredictable, and companies need strong safeguards when using them for important work.

Overall, this week showed that agentic AI is becoming more sophisticated and integrated into mainstream business tools, but companies must proceed with caution and realistic expectations about what these systems can actually accomplish.