Agents Over Bubbles
Agentic AI models now complete multi-hour tasks autonomously, moving beyond simple chat to reliable action.
Tech analyst Ben Thompson, writing from March 2026, argues against the prevailing narrative that the AI boom is a bubble. He identifies three distinct inflection points that demonstrate tangible, escalating progress. The first was the November 2022 launch of OpenAI's ChatGPT, which made the power of large language models (LLMs) accessible to the world, despite early issues with hallucinations and reliability.
The second inflection point arrived in September 2024 with OpenAI's o1 model, which introduced internal reasoning. Unlike auto-regressive models that guess the next token, o1 could self-evaluate its answers before output, dramatically improving reliability and reducing the need for user guidance. This shifted LLMs from being impressive parlor tricks to essential, trustworthy tools.
The third and current inflection point began in late 2025 with the releases of Anthropic's Claude Opus 4.5 and OpenAI's GPT-5.2-Codex. These models, powered by advanced 'harness' software, enabled true 'agentic' workloads. They can autonomously plan and execute complex, multi-hour tasks—like coding projects—with high accuracy, moving AI from a reactive chat interface to a proactive problem-solving partner. This represents a fundamental shift in capability.
- Three distinct AI inflection points: ChatGPT (2022) for accessibility, o1 (2024) for reasoning, and Opus 4.5/Codex (2025) for agentic action.
- Agentic models like Claude Opus 4.5 use specialized 'harness' software to autonomously complete complex, hours-long tasks.
- The progression from chat to reasoning to action demonstrates substantive technological advancement, countering the 'bubble' narrative.
Why It Matters
AI is evolving from a chat-based tool into an autonomous workforce capable of executing complex, real-world projects.