Nikita Kothari

Replacing unreliable “vibe coding” with a rigorous automated evaluation loop using curated datasets, Claude judge agents, and metric tracking for production AI agents.

June 11, 2026

Stop treating AI agents like prompts — treat them like software. To ship in 2026: validated tool contracts, tiered memory, RAG grounding, and deep observability.

May 8, 2026

CUPID shifts focus from rigid SOLID rules to practical, human-centric principles that make code composable, idiomatic, and enjoyable to maintain.

May 8, 2026

Prevent prompt injection in AI agents: default to read-only, require human approval for changes, and authenticate every tool call with end-user zero-trust permissions.

May 6, 2026

Your codebase is essentially a prompt: messy abstractions and "God Classes" pollute the context window, causing AI models to hallucinate or generate bugs.

March 5, 2026

Manual prompt engineering is dead; it is brittle, unscalable, and reliant on "magic strings." DSPy replaces this by treating prompts as optimizable parameters.

March 4, 2026

The industry is shifting from copilots that simply autocomplete code to agentic systems that autonomously plan and execute multi-step workflows in a recursive loop.

February 23, 2026

This article demonstrates a "Threat-Model-First" workflow where we use AI not just to write code, but to aggressively attack our logic before we implement it.

January 30, 2026

This article provides a raw Python implementation, moving beyond high-level frameworks to show exactly how the agentic loop works under the hood.

January 16, 2026