Inside the AI Workflows of Every's Six Engineers
How a six-person engineering team leverages AI to ship like a team three times their size — real tools, real workflows, real results.
The Team Context
Every is a media and software company with a six-person engineering team that builds and maintains multiple products — newsletters, a publishing platform, AI-powered writing tools, and internal systems. They ship features daily, maintain multiple codebases, and operate with startup speed.
What makes their story instructive is the scale mismatch: a team of six doing the work you'd typically need 15-20 engineers for. AI isn't an experiment for them — it's a survival strategy. Every engineer uses AI tools for hours every day, and they've developed refined workflows through months of iteration.
This isn't a 'we tried AI and it was cool' story. It's a 'we redesigned our entire engineering practice around AI' story.
Individual Workflows
Each engineer has evolved their own AI workflow based on their role and preferences:
The Backend Engineer: Uses Claude Code for 80% of backend work. Starts each session by referencing the database schema and API documentation. Relies heavily on multi-file editing for features that span models, services, and API endpoints. Uses headless mode for repetitive tasks like updating multiple endpoints.
The Frontend Engineer: Cursor is the primary tool. Uses Composer for new components and Cmd+K for refinements. Maintains a detailed .cursorrules file with React patterns, styling conventions, and accessibility requirements. Switches to v0.dev for rapid component prototyping.
The Full-Stack Engineer: Alternates between Cursor and Claude Code depending on task complexity. Uses Cursor for quick edits and Claude Code for complex, multi-file features. Runs both tools simultaneously on different branches.
The Infra Engineer: Uses Claude Code with a specialized AGENTS.md that includes AWS patterns, Terraform conventions, and security requirements. AI handles the boilerplate; they focus on architecture decisions and security review.
The Tool Stack
The team's AI tool stack has stabilized after months of experimentation:
Primary Tools:
- Claude Code: Complex backend features, multi-file changes, debugging
- Cursor: Daily coding, frontend work, quick edits
- GitHub Copilot: In-IDE autocomplete for flow-state coding
Secondary Tools:
- v0.dev: Rapid UI component prototyping
- ChatGPT/Claude: Architecture discussions, documentation writing, planning
- Lovable: Quick internal tool prototypes
Key Insight: No single tool covers all use cases. The team's productivity comes from knowing which tool to use for which task and switching fluidly between them. A complex feature might start with architecture discussion in Claude, move to implementation in Claude Code, and finish with UI polish in Cursor.
Lessons Learned
AI Proficiency Is a Skill, Not a Tool
The difference between team members isn't which tool they use — it's how well they prompt, manage context, and review output. Investing in prompt engineering skills pays more dividends than switching tools.
Standardization Beats Individual Optimization
A shared AGENTS.md and consistent conventions across the team produces better results than each engineer optimizing their personal setup. Team-level consistency amplifies individual productivity.
Review Quality Matters More Than Ever
With higher code velocity, review quality becomes the critical bottleneck. The team invested in review checklists, automated checks, and designated review blocks. Faster generation requires more careful review.
Not Everything Should Be AI-Generated
The team learned to identify tasks where AI adds friction: complex architecture decisions, nuanced UX trade-offs, and security-critical code. For these, human-first with AI assist is more effective than AI-first with human review.
Measure the Right Things
Don't measure lines of code or commits per day. Measure features shipped, bug rates, and time-to-production. AI can inflate vanity metrics while degrading real productivity if not applied thoughtfully.
