Frontend after agents
Marc Andreessen predicts programmer, PM, and designer collapse into a single “builder” role. Leading-edge engineers using AI do not work less — they work more and sleep less, because the marginal value of an AI-assisted hour has jumped.
The future-of-frontend-engineering question on the wiki is still open: does frontend remain a distinct specialization over the next three to five years, consolidate into full-stack, or shift toward systems design and infrastructure? What is not open is the direction of pressure. Frontend role compression is already showing up in posting data. AI-native younger hires — people who orchestrate multiple agents in parallel rather than typing every line — are the population Andreessen expects to execute the consolidation first.
His coder-to-builder transition thesis says the three roles collapse because each individual becomes capable of doing the other two via AI. Anecdotal evidence inside a16z points the same way. The day-to-day shape of an AI-native multi-agent workflow — one person whispering to several agents, macro actions over a repo rather than line-by-line edits — is still being documented. Karpathy and Conviction portfolio engineers independently describe that shift from December 2025 onward.
Kareem Amin, who runs Clay, describes a different non-engineer getting the same kind of leverage. Not Andreessen’s collapsed job title. Sales. “We started with the really abstract ambition of how do we give the power of programming to more people.” The product is coding-shaped — composable go-to-market primitives, arbitrary integrations — pointed at outbound. He says the company was already taking off before ChatGPT, and that language models then “boosted everything” because the integration surface was already there. It is a podcast claim, not a revenue case study, and not a Clay valuation. The host’s “over $4 billion” is not filed as fact.
Productivity ramps roughly 20×, but instead of working less they work more, sleep less, and command higher comp, because the marginal value of an hour of fluent AI-assisted work has gone up sharply.
Marc Andreessen, AI vampire pattern framing
Andreessen calls it the AI vampire pattern. Jevons paradox in software: if the barrier to building falls, demand for software rises. Bank tellers did not disappear when ATMs arrived; cheaper branches meant more branches. The wiki holds that framing alongside frontend compression without resolving whether net frontend headcount grows or shrinks. Both can be partially true — fewer generic FE seats, more builders who ship end-to-end.