🧠 Food for thought

Implications, context, and why it matters.

Meta’s ad engine runs on a three-part AI system

• Meta’s AI ad system now works as a tightly linked three-part setup that changes how ads reach people 1.
• First comes Andromeda, an internal ranking system that runs on NVIDIA chips and scans millions of ads to build a candidate pool for each user 1.
• Then the Generative Ads Model, or GEM, acts as the central model that predicts results and lifted Instagram conversions by up to 5% 2.
• Lattice makes the final pick for which ad wins, with conversion gains of up to 6% 3.
• That stack may help explain higher average ad prices by making each placement work harder.

Advertisers give up control as Meta hands more work to AI

• For advertisers using these tools, success now depends less on manual targeting and more on a steady flow of fresh creative ideas because the system favors variety 1.
• As automation spreads, marketing teams get less visibility into delivery while Meta pushes broader audiences and gives its systems more say over who sees an ad 2.
• Those efficiency gains support a huge spending plan, with Meta projecting 2026 capital expenditures of US$125 billion to US$145 billion 4.
• Meta is cutting jobs while it trains AI agents by tracking employee workflows, a sign that software may take over more internal tasks 5.