I am going to miss coding

Agentic AI moves beyond passive code suggestions by letting the model read, edit, and commit code directly in your workspace. The author uses Cursor, a VS Code-based tool that builds embeddings of an entire repository so the assistant can answer requests with full context. This eliminates the back-and-forth of copy-pasting snippets and lets you ask for concrete changes like adding a delete button or generating a full feature with tests.

The piece also warns that giving an external service access to your code creates real privacy and IP risks. Embeddings obscure source data, but companies may still need to audit what leaves the premises, especially when customer PII is involved. Cursor offers a privacy mode to mitigate these concerns, showing how tools are adapting to corporate compliance.

Practical advice focuses on controlling the agent: keep prompts specific, review each commit, and use source control to roll back unwanted changes. The author suggests treating the AI as a junior programmer you supervise, asking it to draft PRDs or user stories first, then incrementally applying and verifying the code. This approach balances speed gains with the danger of hallucinated or off-target modifications.

For technical leaders, the takeaway is clear: agentic coding can boost team productivity, but it requires new guardrails-policy around data exposure, rigorous code review, and a culture that treats AI output as assistive rather than authoritative.