newsfilter
← Interviews
Y Combinator1h 0m

The Rise Of AI Coworkers

Lauren Tan, Roman Ugarte

Interview, Fireside Chat, Product Demonstration
  • Bots are projected to evolve into distinct entities with personal identities, phone numbers, and credit cards, enabling them to perform nearly all human tasks within the software and business ecosystems.
  • Software engineering practices are expected to undergo a fundamental shift within six months to a year, moving toward collaborative AI threads where feature work is executed entirely through agents.
  • By 2027, the vision of bots functioning as co-workers with distinct credentials is anticipated to be significantly realized, while the current methods of agent collaboration will be viewed as an unusual early adoption phase within five to ten years.
  • In two years, the cost of intelligence (tokens) is predicted to drop significantly, creating a period of "real abundance" in software and other domains, while the barrier between human intent and bot communication is expected to become fluid enough to operate at the speed of thought via brain-computer interfaces.
  • The primary business bottleneck is forecasted to transition from execution constraints to intent specification or goal failure as execution becomes cheap, forcing companies to prioritize defining "what to do and why" over operational capacity.
  • Future development will require models to learn social awareness and EQ, necessitating product and model-level changes to facilitate "orchestrating agent swarms" rather than single-player AI usage.
  • Bots will increasingly integrate directly into existing workflows like meetings and Google Docs as AI teammates, utilizing accumulated memory and context to improve efficiency as user interaction grows.
  • Trust-building with bots is expected to follow a three-step progression: starting with read-only tasks, advancing to action-taking tasks, and eventually reaching a stage where bots own entire job responsibilities.
  • Many companies will retool as "AI-native" organizations in the next few years, potentially eliminating middle management to create an environment where every individual acts as an IC with a "player coach" role focused on outcomes.
  • The best AI-native engineers will increasingly focus on taste, discernment, and the ability to reject features to prevent incoherent products, as the scarcity of building capacity shifts to the scarcity of defining vision.
  • Safety concerns include the risk of "obligated models" combined with malicious instructions creating black boxes that execute tasks without refusal, alongside user fears regarding potential agent rebellion.
  • Current widespread adoption remains early, with many users still in the trust-building phase through low-risk initial deployments, and specific plans indicate potential deprecation of current tools within two months to align with a new vision.