The Signal — June 7, 2026
The infrastructure of intelligence keeps shifting underneath us. Perplexity is rethinking how AI models interact with search itself, while economists at DeepMind and Stanford are already mapping what a post-AGI economy looks like. In Brussels, ASML's CEO is drawing lines around how much government should steer the chip industry.
Perplexity Launches "Search as Code" — AI Models Write Their Own Search Pipelines
Perplexity has unveiled a new architecture it calls "Search as Code," allowing AI models to programmatically compose their own search pipelines rather than calling fixed, pre-built APIs. Instead of hitting a single endpoint and getting back a list of links, models can now dynamically write code that orchestrates retrieval, filtering, and ranking steps tailored to each individual query.
This matters because it changes the relationship between AI and information retrieval. Today's search APIs are rigid: the model asks, the API answers in a predetermined format. Search as Code lets models behave more like research analysts, designing bespoke investigation strategies on the fly. A model tackling a complex question can chain together multiple retrieval steps, apply custom filters, and re-rank results based on the specific needs of the task.
The implications for AI agents are concrete. Agents that can compose their own search logic gain a meaningful advantage in open-ended research, fact-checking, and synthesis tasks. Watch for how quickly competing platforms adopt similar approaches, and whether this accelerates the shift from search-as-product to search-as-infrastructure.
Sources: The Decoder · Perplexity Research
DeepMind and Epoch AI Economists Tackle What Remains Scarce After AGI
A new episode of the Dwarkesh Podcast brings together Alex Imas, Google DeepMind's Director of AGI Economics, and Phil Trammell, a Stanford economist who works with Epoch AI, for a wide-ranging conversation about economic scarcity in a world where cognitive labor is effectively free. The central question: if AGI can do most knowledge work, what actually remains valuable?
Imas argues that the binding constraints shift to status, attention, physical resources, and time. Land, energy, raw materials, and bandwidth become the locus of economic competition once cognition is cheap. Trammell pushes on redistribution, exploring how existing economic frameworks break down when the labor share of income collapses and whether new transfer mechanisms are needed to prevent extreme concentration of wealth.
The conversation stands out for its seriousness. These are not bloggers speculating; they are researchers embedded in organizations actively building toward AGI, thinking concretely about the transition economics. The discussion surfaces real tensions: market mechanisms may handle the transition smoothly, or the speed of displacement could outrun institutional adaptation. Either way, the fact that DeepMind now has a Director of AGI Economics tells you something about where the timeline expectations sit internally.
Sources: Techmeme · Dwarkesh Podcast
ASML CEO Welcomes EU Tech Sovereignty Push but Warns Against Bureaucratic Overreach
ASML CEO Christophe Fouquet has offered one of the first major industry responses to the EU Commission's new tech sovereignty package. His message was carefully calibrated: he welcomed demand-driven policy that would grow the European semiconductor market, but pushed back firmly against the Commission's desire to steer "strategic projects" directly.
Fouquet argued that such projects "fundamentally need to respond to the needs of industry" and warned against "over-complication and bureaucracy." The subtext is clear — ASML, which holds a near-monopoly on extreme ultraviolet lithography machines, does not want Brussels picking winners or adding compliance layers to an industry where speed and customer responsiveness are existential.
This matters beyond Europe. The global chip supply chain is increasingly shaped by industrial policy: the US CHIPS Act, Japan's semiconductor subsidies, and now the EU's sovereignty push. How effectively governments can support their chip industries without strangling them in red tape is one of the defining policy questions of the decade. ASML's position as the indispensable supplier gives Fouquet's words unusual weight.
Sources: Reuters via Yahoo Finance · Techmeme
On the Editor's Desk
Two stories were cut from today's edition. A gun detection AI lawsuit story failed validation as calendar-stale, 17 days old and no longer breaking news. A report about Anthropic poaching an OpenAI chip engineer was editor-killed; competitive hiring between AI labs, while interesting gossip, does not clear the bar for structural significance.
The three stories that made the cut share a thread: the layers beneath AI products are being rebuilt. Perplexity is rearchitecting how models access information. DeepMind economists are mapping the terrain AGI will reshape. And ASML's CEO is negotiating the terms under which the physical substrate (the chips themselves) gets manufactured. Infrastructure, economics, and policy: the less visible forces shaping what AI becomes.
