Ori Goshen, Co-founder & Co-CEO of AI21 Labs: AI - All The Work!

Company Overview & Market Context

• AI21 was founded seven years ago by co-founders Ori Goshen (Co-CEO), Yoav Shoham (Stanford professor), and Amnon Shashua (former Hebrew University professor).
• The company operates as a frontier model builder focused specifically on the enterprise sector.
• Knowledge work now constitutes approximately 76% of the U.S. economy, up from 15% in 1959 when the term was coined by Peter Drucker.
• Global work is defined by the speaker as comprising roughly six billion distinct tasks.

Technical Challenges in Agentic AI

• LLMs achieve high accuracy (~90%) on single-step tasks but fail on multi-step workflows due to compounding errors.
• Traditional solutions involve rigid static chains requiring pre-programmed steps, which are time-consuming and inflexible.
• Simple prompting lacks the consistency required to guarantee that agents follow all necessary steps in complex workflows.

Product Announcement: "Maestro"

• AI21 introduced "Maestro," a model-agnostic AI agent platform designed to automate mission-critical enterprise workflows.
• Key Features:
  • Allows developers to explicitly define requirements, instructions, and tools for validation.
  • Provides control over computational budgets (time and cost) for specific tasks.
  • Generates multiple execution plans to determine the optimal path based on available resources.
  • Offers "Maestro Validated Output," ensuring errors do not compound through step-by-step verification.
  • Delivers an execution graph for full visibility into agent reasoning and actions.
  • Includes a "report card" feature showing requirement satisfaction and system confidence levels.

Performance & Deployment

• Benchmarks indicate that on-the-shelf models boosted with Maestro see dramatic increases in accuracy levels.
• AI21 is "dogfooding" the platform internally, using Maestro to build and deploy reliable agents for enterprise clients.

Strategic Vision: "Super Productivity"

• The speaker rejects binary utopian/dystopian narratives regarding AI's impact on employment.
• Case Studies Cited:
  • Microsoft reduced headcount after generating 30% of code with AI, illustrating workforce optimization.
  • Uplovin doubled revenue (from $3.6M to $7.6M) with flat headcount by utilizing AI, demonstrating efficiency gains.
  • The video content industry expanded from 1 million creators in 2000 to 64 million, illustrating an explosion of human capacity.
• Core Philosophy: AI reduces the gap between ideation and execution, enabling humans to pursue underserved needs and create new market formats (e.g., ASMR, unboxings) rather than resting.
• Definition: "Super productivity" is the process where AI empowers humans to execute faster and solve more problems, not to replace intelligence.
• The speaker posits that automating just 10% of company tasks can unlock previously unconsidered opportunities.