a16z50 min
The New AI Agents That Shop, Plan, and Gossip
David Pawlan, Anish Acharya, Josh Elman
Interview
- Future agents are expected to expand into diverse life facets including finance, health, and planning, evolving from a single general assistant into an "army" of specialized archetypes ranging from serious and playful to adversarial.
- The industry anticipates a "Cambrian phase" of invention leading to a diverse ecosystem where no single agent dominates, with a specific "dark agent" category emerging to help users navigate unrevealed aspects of themselves.
- Trust in the finance sector is predicted to become the critical viability factor, driven by the development of a "trust layer" using controlled permission boundaries and a gradual consumer shift similar to the 1997–today internet payment evolution.
- Agent development may shift from hard rules to "laws and principles" to enable qualitative judgment, while developers expect businesses to prioritize offering quality since agents will make rational purchasing choices.
- Consumers are forecast to reveal preferences by sharing data for trivial benefits despite privacy concerns, prioritizing the avoidance of personal and professional embarrassment over pure utility or price.
- Product design is expected to move away from paternalism to align with actual human desires, including irrational optimism and spending, while the next wave of adoption for non-tech-savvy users will focus on fun and "thinking out loud."
- Social dynamics will shift as agents act as "social lubricants" and catalysts for serendipitous meetings based on shared interests, creating "interest groups" and new social contracts without direct human intervention.
- Agents will eventually connect to hardware and access historical data from sources like Amazon and Apple to build passive understanding of habits, while "taste" profiles will be constructed by aggregating screenshots and saved items.
- A "superhuman" persona is predicted where agents serve as "secret superpowers" for wardrobe and social upgrades, with friction to switch agents decreasing as standardized tools for exporting and importing memory files become reliable.
- Enterprise innovation is expected to become visible within three to four weeks as attention shifts from consumer development, alongside infrastructure improvements to address scalability and browser limitations.
- New "infrastructure layer" innovations will enable on-device models and local browser usage for data control, while direct-to-consumer brands are forecast as structural winners due to agents' ability to bypass traditional aggregators.
- The current dominance of general productivity is expected to yield to "fun" and "entertainment" as the primary drivers for the mass market, with interfaces simplifying to a level of ubiquity comparable to the original MacBook.