How Ex-Google Founders See Opportunity in Agents and RL

Editor’s Take · AI Serial Entrepreneur (Content distilled by AI; views belong to the original author; no need to read the full source)

The former Google DeepMind founder reveals that Gemini 3 turns things around by leveraging TPU and data synergy; Agent development must first solve cloud provider compatibility and long-tail gaps. For those chasing profit, vertical Agent tool layers and RL environment services are the current tailwind, but pure model-layer involution should be avoided. First step: pick one high-frequency office scenario, connect Claude or Gemini via existing cloud discounts, and build a “trainable tool layer” to validate user retention.

• Vertical Agent opportunities: fill the gap where foundation models fall short in long-tail office scenarios
• RL startup direction: offer RL environments or RL as a Service
• Pitfall to avoid: Agent developers should confirm cloud provider compatibility before choosing a model
• Tool-layer opportunities: boost execution efficiency by using code-driven tools (e.g., PTC)
• Validation cost: use cloud provider discounts to lower early-stage Agent deployment costs

Original article · LateTalk: Read original →