From chatbots to collaborators: How AI agents are reshaping enterprise work

Scott White is still afraid of the sooner where artificial intelligence becomes a real business partner from an innovation. A year ago, the product brought by the product Claude ai horse Anthropical Early AI coding tools followed each other as they could complete a code line. Today, the production-finished software itself, despite being a professional programmer.

“I don’t think about my work while trying to write a prd and convince someone to do something” in white fire extinguisher VB Transform 2025Venturebeat’s annual enterprise AI summit in San Francisco. “The first thing I do, I can build a prototype that can work on this on our structure server, and then I can actually share a demo that works.”

This turn represents a broader transformation in the adoption of enterprises in the adoption of enterprises, and moves outside simple conversations that meet the complex “Agentic” systems capable of autonomous work. The experience of the White offers an idea of ​​what can happen to millions of other knowledge employees.

Following the autonomous programming code: AI’s Breakneck evolution

The evolution was very fast. When the white anthropic, the company combines Clod 2 The model can handle the completion of the main text. Release Claude 3.5 Sonnet Allowed all applications that caused properties such as Artifacts allows users to create a special interface. Now, with Clod 4 Collect 72.5% on Swe-bench coding benchmarkThe model can work as a “full remote agent” as a software engineer, called white.

Clamping codeThe company’s latest coding tools can analyze all coding, all codes, search for the Internet, responds to code review reviews and responds for solutions – hours asynchronous. White, Claude code was written by 90% of the AI ​​system.

“This is like the whole agency process in the background that is not possible six months ago,” he said.

Enterprise giants work with AI agents from weeks to a few minutes

The effects are far from software development. Novo NordiskThe Danish pharmaceutical giant has previously completed the same work in 10 minutes and combined previously 10 weeks to end working streams to end clinical reports. Gitlab Uses technology for everything from sales offers to technical documents. Intuit places the cloan to provide tax recommendations directly to consumers.

The EU differs between different levels of integration: simple language models that answer questions, web search, ai-nip tools that include structured work flow and multi-tool and iterative key, models such as all agents that can be used maximum use.

“I think like something that is a goal about an agent and then you can do a lot to achieve this goal,” he said. The main Enabler is what it is called “impassable” ties between model intelligence and new product capabilities.

Infrastructure Revolution: Construction networks of AI employees

Has been developing a critical infrastructure Anthropic model context protocol (MCP), white, described as “USB-C” for integration. Each source of data or tools builds separate connections, provides a standard way for MCP, AI systems, offers a standard way for AI systems.

“It’s the ability to learn really,” white, “it can be distributed through the open source source protocol of the integrations built by a company that can be used.

For organizations that want to implement AI agents, it recommends that white small and growth. “Do not try to build a system of agency from scratch,” he said. “Set up the component of this, make sure the component works, then set up a component.”

EU agents also stressed the importance of evaluation systems to ensure the implementation of the implementation. “The world is a new prd”, “Touching on white, product requirements, stressing how new methods should develop new methods to evaluate the EU performance on specific work assignments.

From AI Assistants to AI: The next workforce border

The White Assembly, which is looking forward, is accessible to non-technical workers, similar to how to develop ai development coding opportunities. Only one AI agent of individual persons imagine a future that does not control all organizations of specialized AI systems.

“How can everyone be your own mini CPO or CEO?” He asked the white. “I don’t know exactly what it looks like but I woke up and I want to be there.”

The turning white descriptions reflect more extensive industrial trends as the company’s eclipse with the expanded capacity of AI. Although early adoption is directed to experimental use, enterprises are increasingly inhabited by the basic work processes by changing how the work is developed.

When AI agents are more autonomous and competent, the problem is to carry out the instructions to manage tutorial machines for a long time to manage independent AI employees. This future for white comes – a production feature at a time.