Today, agent practice has cycled through vibe coding, harness, loop, and graph. How should we make sense of this? The Way of Agentic is about cracking the code beneath these cycles and finding the most fundamental principles. I shared this topic at the GTLC Hangzhou 10th-anniversary station to unanimous praise, and hope to share my views in this lightning talk.
The Way of Agentic, simply put, upgrades AI from a “temporary Q&A tool” into a “long-term execution system”. It emphasizes four things: clarify goals, accumulate context, set constraints, and plan before acting — writing every experience back into the system to compound.
Its main line can be summarized as: Code → Context → Agent Plan → Business Brain → OPC (one-person company). Context matters most: instead of re-explaining every time, project rules, historical experience, test commands, architectural conventions and risk boundaries are accumulated long-term for agents to call on repeatedly.
In practice, the Way of Agentic does not mean “spinning up more agents”. It means: state the goal clearly; write context into documents and rules; turn constraints into hard guardrails; have AI plan first, then act; verify results with tests, builds, screenshots and reviews; and write experience back into the system to avoid repeating mistakes.
It also emphasizes a closed loop: every fix, review, and pitfall should become a long-term capability of the system rather than staying in a single conversation — write solutions into documents, turn common issues into lint rules, and distill experience into skills or retrospective records.
In one sentence: the Way of Agentic = using goals, context, constraints, planning, verification and experience write-back to upgrade AI from a one-off assistant into an execution system that continuously accumulates capability.