Inside the AI Workflows of Every's Six Engineers

Rhea Purohit interviews six engineers at Every about their personalized AI stacks. Each engineer runs a different configuration of tools, models, and workflows to build and maintain four AI products (Sparkle, Cora, Spiral, Monologue), a consulting business, and a daily newsletter.

Yash Poojary — General Manager, Sparkle

• Mac Studio + laptop. Runs Claude Code on one machine, Codex on the other, feeding them identical prompts to compare results.
• Claude Code: "friendly developer" (breaks things down). Codex: "technical developer" (literal, precise).
• Uses Figma MCP so Claude reads design system directly (colors, spacing, components), not screenshots.
• Warp terminal.
• Maintains a "learnings doc" — two lines per push about what he learned, stored in the cloud as rolling memory for AI tools.
• Built AgentWatch, an app that pings him when a Claude Code session finishes so he can run multiple sessions in parallel.
• Structures day around one big task plus smaller background ones. "The problem with CLIs is it's easy to get derailed and lose focus."
• Mornings: focused execution. Afternoons: exploration with new agents, apps, features.

Kieran Klaassen — General Manager, Cora

• Everything starts with a plan generated in Claude Code using custom agents and workflows.
• Three plan sizes: small (one-shottable), medium (span a few files, include review), large (require manual typing, research, iteration).
• Uses Context 7 MCP (by Upstash) to pull up-to-date, version-specific docs and code examples into prompts.
• Plan → GitHub → "work command" turns plan into coding tasks.
• Claude Code primary. Occasionally Codex or Amp for "nerdier" features.
• Review command runs after work: Claude leads, but also uses Cursor and Charlie.
• Process loops until feature is ready to ship.

Danny Aziz — General Manager, Spiral

• Primary tool: Droid (Factory's CLI). Runs Anthropic and OpenAI models side by side. ~70% of work here.
• Uses GPT-5 Codex for big feature builds, then Anthropic models for refinement and detail work.
• Talks through implementation plans with GPT-5 Codex, asking about second- and third-order consequences.
• Mostly single monitor or laptop. Adds second desktop only for deep design implementation (Figma side-by-side with build).
• "I don't use Cursor anymore. I haven't opened it in months."
• Main interface: Warp (split screens for fast task-switching). Uses Zed for reviewing plan files and specific code.

Naveen Naidu — General Manager, Monologue

• Everything starts in Linear. "If it's not in Linear, it doesn't exist." Every ticket links back to its original source.
• Recently migrated from Claude Code to Codex for daily work.
• Two-track execution:
• Codex Cloud: brainstorms approaches, generates draft PRs (not to merge, but to explore ideas and surface edge cases). Accessible via iOS ChatGPT app or web for async background tasks.
• Codex CLI: real build, refining plan.md, driving file edits step by step in Ghostty.
• Other tools: Xcode for native macOS, Cursor for backend. MCP integrations with Linear, Figma, and Sentry.
• Review process: Codex /review command for automated bug scanning, manual side-by-side before/after comparison, Sentry error logs check for bug fixes.
• Uses Monologue (speech-to-text app he built) to dictate prompts, write ticket descriptions, and update plans.

Andrey Galko — Engineering Lead

• Doesn't chase every new tool. Longtime Cursor user who left due to pricing (hit monthly limits in a week).
• Switched to Codex, credits GPT-5's improved code generation.
• Earlier OpenAI models produced "suboptimal code" — technically functional but inconsistent, "lazy." GPT-4.5 and GPT-5 changed that.
• Codex strengths: always good at non-visual logic. With GPT-5-Codex, "it finally got good at the user interface, too."
• "I applaud the people at OpenAI for becoming a real menace to Anthropic's code generation reign."

Nityesh Agarwal — Engineer, Cora

• Entire stack on a MacBook Air M1. "I'm the kind of developer who doesn't like changing my tools often."
• Primary tool: Claude Code (Max plan) for all AI-assisted coding.
• Spends hours researching codebase and sketching detailed plans with Claude before writing code.
• Stays in a single terminal, giving "100 percent attention to the one thing that Claude is working on."
• Watches Claude's work "like a hawk," finger on the Escape key.
• Has shortened Claude's leash, interrupting mid-process to ask for explanations — reduces hallucinations, sharpens his own skills.
• On Claude dependency: "I realize that I've placed too much of my trust in Anthropic, which leaves me vulnerable." When Claude glitched for two days, alternatives didn't match.
• GitHub as review interface: team reviews PRs, leaves line-by-line comments, Claude Code fetches and reads those comments to make fixes collaboratively.
• Cursor and Warp are "solid nice-to-haves."

Cross-Cutting Themes

All Tools Mentioned

AI Coding Agents: Claude Code, Codex (OpenAI), Codex Cloud, Codex CLI, Amp, Cursor, Charlie, Droid (Factory) Terminals/Editors: Warp, Ghostty, Zed, Xcode MCP Integrations: Figma MCP, Context 7 MCP, Linear MCP, Sentry MCP Other: Linear, GitHub, Figma, Sentry, AgentWatch, Monologue, Sparkle Products built by Every: Sparkle, Cora, Spiral, Monologue