The software industry is pricing in its own obsolescence. Following two years of treating AI as a conversational partner, the market now sees its arrival as an autonomous workforce. The S&P 500 Software Industry Index dropped 20% as investors realized AI coding agents are moving from proof-of-concept to production, directly replacing traditional engineering workflows.
The shift is from chatbots to agents. As Anthropic co-founder Jack Clark explained on *The Ezra Klein Show*, an agent takes a command and works independently over time, using tools to complete complex tasks. He described building a multi-system species simulation in ten minutes - a job that would take a human engineer days.
Jack Clark, The Ezra Klein Show:
- An agent is something where you can give it some instruction and it goes away and does stuff for you, kind of like working with a colleague.
- The way that I think of these systems now is that they're like little troublesome genies that I can give instructions to, and they'll go and do things for me.
This new capability stems from a change in training. Modern 'reasoning' models aren't just predicting text; they're trained in active environments with spreadsheets and compilers, learning through trial and error. This lets an agent course-correct mid-task - like switching search strategies when a paper isn't in an expected archive - without human intervention.
Anthropic is aggressively capitalizing on this shift. By making coding its core competency, it has turned a technical niche into an enterprise gateway. David Sacks argued on *All-In* that this was a calculated bet: if a model can write its own code, it can recursively build its own future. The strategy is working; Anthropic reportedly added $6 billion to its annual run rate in February alone.
Meanwhile, OpenAI is consolidating. It shelved plans for an 'adult mode' after technical and safety failures, refocusing on coding and enterprise sales. The labs are running different races: OpenAI dominates consumer subscriptions, while Anthropic owns the developer API market. As Chamath Palihapitiya noted, their revenue streams are near opposites, making them parallel giants for now.
The transition is messy. Users often fail by treating agents like intuitive humans. Clark stresses they are literal-minded genies; vague prompts yield buggy messes. Professional results require users to become architects, writing exhaustive specification documents for the agent to follow precisely.
This evolution is rewriting the knowledge labor market. The initial 20% stock drop is just the market's first reaction to a fundamental change: AI is no longer a tool for engineers, but a replacement for the engineering process itself. The question is how far the dominos fall.
