AI projects around the world that show what AI can do in practice

The best AI projects have one thing in common, and it is not the model they used: they solve one clear task with a measurable result. Here are five examples and what to take from them.

Klarna: an AI assistant in customer support

In February 2024 the Swedish payments company Klarna said its AI assistant handled 2.3 million conversations in its first month, about two thirds of customer service chats. According to the company it did the work of 700 full-time agents and cut resolution time from 11 to under 2 minutes. Klarna later stressed that customers must always be able to reach a person.

Lesson: support is an ideal first process, as long as the agent answers from verified documents and can hand cases to a person.

Morgan Stanley: company knowledge in one click

In 2023 Morgan Stanley gave its financial advisors an assistant built on GPT-4 that answers from a large internal library of research and procedures. Advisors no longer dig through documents; they get an answer with its source.

Lesson: a knowledge base (RAG) over your own documents is a fast route to value, without training your own model.

AlphaFold: AI that won a Nobel Prize

Google DeepMind’s AlphaFold predicts the three-dimensional structure of proteins. For this work Demis Hassabis and John Jumper shared the 2024 Nobel Prize in Chemistry with David Baker. The database of predicted structures is freely available to scientists worldwide.

Lesson: AI changes fields most where there is plenty of data and a clearly measurable result.

Google Gemini: an agent as a colleague with its own account

In October 2026 Google launched an agent in Gemini for businesses that completes tasks across email, documents and company systems. It gets its own work account, and every step is written to an audit trail.

Lesson: an agent needs an identity, permissions and a record of its steps, just like a new hire.

Fujitsu: agents for retail

In October 2026 Fujitsu announced a trial environment for four retail agents: sales analysis, loyalty analysis, merchandise planning and store-manager support. A full launch is planned for 2027.

Lesson: even large companies start with a set of small agents with clear roles, not one do-everything agent.

What to take away

• Start with one process and a measurable result.
• The agent should answer from your data and be able to hand work to a person.
• Set permissions and a record of every step from day one.