This week brought major advances in agentic AI - artificial intelligence systems that can work independently to complete tasks. IBM released CUGA, an open-source agent designed to automate complex business workflows by orchestrating tasks, connecting to APIs, and generating code. The agent succeeded in completing more than half of its assigned tasks in benchmark tests, with a 61.7% success rate on WebArena and 48.2% on AppWorld.

Google also unveiled a reimagined Gemini Deep Research agent powered by their latest Gemini 3 Pro model. This agent does more than just generate reports - it can now be embedded into other applications through Google's new Interactions API, letting developers add research capabilities to their own apps. Google claims the agent has improved accuracy and reduced hallucinations, or false information.

However, not all companies are seeing strong returns from their AI investments yet. A Reuters report found that many businesses are struggling to see real value from generative AI, despite widespread adoption. Major business leaders noted problems like AI systems being inconsistent, failing at seemingly simple tasks, and struggling with long documents.

Notion, a popular productivity tool, is expanding into an "everything app" with new AI features including enterprise search and a knowledge work agent that can perform extended tasks. This shift shows how artificial intelligence is moving from simple tools into more sophisticated systems that can do real work for companies.

Experts are warning that agentic automation should be treated with caution. Strong quality checks and safety guidelines are essential, especially for customer-facing tasks that could create errors or compliance problems.