Scott White is still delighted how quickly artificial intelligence has transformed from latest into a real working partner. Just over a yr ago it runs the product Claude AI On Anthropic I watched the early AI coding tools could barely complement one code line. Today he builds a ready-made program for production-if he is not a skilled programmer.
“I do not think about my work as a prd and I’m trying to persuade someone to do VB Transform 2025Annual peak AI Enterprise AI Venturebeat in San Francisco. “The first thing I do is that I can build a feasible prototype on our staging server, and then share its demo actually works.”
This change is a broader transformation in how enterprises accept artificial intelligence, going beyond easy chatbots, which answer questions to sophisticated “agency” systems capable of autonomous work. White experience offers insight into what can come for tens of millions of other knowledge employees.
From the end of the code to autonomous programming: Breakneck AI evolution
Evolution was extremely fast. When White joined Anthropik, corporations Claude 2 The model could handle the basic ending of the text. Issue SONET CLAUDE 3.5 allowed the creation of entire applications, leading to such functions Artifacts which permit users to generate custom interfaces. Now with Claude 4 Achieving a result of 72.5% on Swe-Bench Coding BenchmarkThe model can function as what White calls “a fully remote agency software engineer.”
Claude codeThe latest company coding tool can analyze entire code databases, search the Internet for the API interface, demanding problems, respond to comments to the code review and items solution – all during asynchronous work. White noticed that 90% of the Claude code itself was written by the AI system.
“It is like the whole agency process in the background, which was not possible six months ago,” explained White.
Giants Enterprise lowers working time from weeks to minutes with AI agents
Implications go far beyond the development of software. Novo NordiskThe Danish pharmaceutical giant, integrated Claude with work flows, which previously took 10 weeks to full clinical reports, ending the same work in 10 minutes. Gitlab He uses this technology for every little thing, from sales proposals to technical documentation. Intuit implements Claude to give tax advice directly to consumers.
White distinguishes various levels of integration of artificial intelligence: easy language models that answer questions, models reinforced with tools reminiscent of website search, structural work flows that contain artificial intelligence for business processes, and full agents that may achieve goals autonomously using many tools and iterative reasoning.
“I think about the agent as something that has a goal, and then he can simply do a lot of things to achieve this goal,” said White. The key switching factor was what he calls the “inexorable” relationship between model intelligence and latest product capabilities.
Infrastructure Revolution: building a network of collaborators AI
The critical development of infrastructure was Contextual protocol of the Anthropic model (MCP), which White describes as “USB-C for integration.” Instead of corporations building separate connections with each data source or tool, MCP provides a normalized method of AI systems to access corporate software, from Salesforce to internal knowledge repositories.
“It really democratizes access to data,” said White, noting that the integrations built by one company might be made available and reused by others through the Open Source protocol.
For organizations that want to implement AI agents, White recommends starting small and incremental building. “Do not try to build the entire agency system from scratch,” he advised. “Build his component, make sure the component works and then build the next component.”
He also emphasized the importance of assessment systems to ensure the operation of AI agents as intended. “Evals is a new PRD,” said White, referring to product requirements documents, emphasizing how corporations must develop latest methods of assessing AI results in the scope of specific business tasks.
From AI assistants to AI organizations: the next limit of the workforce
Looking to the future, White predicts that the development of artificial intelligence becomes available to non -technical employees, in addition to coding proceedings. He imagines the future in which individuals manage not only one AI agent, but all organizations of specialized AI systems.
“How can everyone be their own mini CPO or CEO?” White asked. “I don’t know what it looks like, but it’s something that I wake up and I want to get there.”
White transformation describes wider industry trends because corporations are struggling with the developing AI capabilities. While early adoption focused on experimental cases of use, enterprises are increasingly integrating artificial intelligence with basic business processes, they typically change the way of work.
Since AI agents develop into more autonomous and talented, the challenge moves from teaching machines to perform tasks for managing AI colleagues who can work independently for a very long time. For White, this future is coming – one production function at once.