Accuracy and Agentization.

• Without high accuracy, agentization is difficult. However, accuracy issues aren’t confined to technology. They occur at various phases of interactions, some happening at the user-side (e.g., human language limitations and ambiguous user instructions) and some happening at the AI-side (e.g., hallucinations)
• Why should AI initiate conversations or ask questions to improve input accuracy?
  • There’s a paradox where users need coding-level language precision to convey their intentions accurately. To address this, AI should ask clarifying questions.
  • OpenAI’s ChatGPT 4.5 Deep Research, one of the more cutting-edge feature, always asks clarification first before zeroing-on its research topic. From a user experience perspective, this feels much more useful. From the design side, this small interaction increases the accuracy of the input. From the server-side, this approach reduces computing resource waste.

Personal Experience with Major AI Services

• As someone who uses almost all major AI services on a paid basis, I interact with them more than with people.
• Recent “wow” moments include Perplexity’s integration of reasoning model like DeepSeek in its search feature and ChatGPT 4.5’s Deep Research.
• The latter feels like delegating tasks to a real assistant. Is this the first experience of agentization? $200 per month might seem steep, but it’s incredibly cheap compared to hiring a person—though it remains a passive tool.
• Despite its superior performance compared to similar features from other companies, the $200 Pro plan’s 120-session limit must mean the feature is extremely resource-intensive. The feature is available for $20 Plus users, so definitely check it out. (I’ve heard good things about Gemini 2.5 Pro and plan to try it out.)