← Work

Product

Cortex Code

Cortex Code is Snowflake’s AI-assisted CLI that
embeds intelligence into existing developer
workflows—helping developers write, run, and debug
with AI without leaving the terminal.

Details

Role
Lead product designer
Team
1 Designer · 1 PM · 4 Engineers
Scope
CLI experience, AI interaction patterns, developer workflows
Users
Developers working across code, terminal, and AI tools

Problem

Core problem

AI capabilities existed—but were not integrated
into how developers actually work. Frequent context
switching between IDEs, terminals, and separate
AI tools created friction and slowed iteration.

Solution

Design for workflow-native AI

Embedding intelligence into the CLI
where developers already operate—
making AI behavior visible, keeping
developers in control, and connecting
prompting, execution, and feedback
into a single loop.

Result

Adoption + confidence

Reduced friction integrating AI into
daily workflows. Developers adopted
Cortex Code with more confidence—
faster iteration, more efficient
debugging, and less hesitation around
AI-assisted commands.

Context

Developers work within IDE-driven workflows—code editing in VS Code and similar tools, with terminal/CLI for execution. As AI capabilities expanded, the product needed to evolve from a standalone feature into an integrated part of how developers already work.

The system broke because…

  • AI lived outside the workflowdevelopers context-switched constantly
  • Opaque AI behaviorlow trust in outputs
  • Automation without controldevelopers hesitated to use AI

Strategy

Design for workflow-native intelligence.

Connecting prompting, execution, and feedback into a single loop.

  1. Prompt

    Natural language

  2. Generate

    Commands

    Suggestions

  3. Review

    Edit

    Validate

  4. Execute

    CLI

    Output

  5. Iterate

    Debug

    Refine

Decision 1

Bring AI into the CLI.

Insight

Developers don’t want to leave their workflow to use AI.

Tension

UI-based AI → discoverable, easier to guide vs. CLI-based AI → integrated into workflow, but more constrained in design.

Decision

Brought AI into the CLI instead of expanding the UI.

Trade-off

Reduced discoverability and guidance compared to a UI-based experience.

Cortex Code welcome screen in the terminal with theme selection

Decision 2

Make AI behavior visible to developers.

Insight

Developers need to understand what the AI is doing to trust it.

Tension

Abstract AI → simpler, but opaque vs. visible AI behavior → more trust, but more noise.

Decision

Made AI behavior visible—surfacing commands, reasoning, and outputs before execution.

Trade-off

Surfaced commands and outputs that required interpretation before acting.

Cortex Code surfacing troubleshooting steps, bash output, and error tracebacks in the terminal
Cortex Code showing file reads, agent reasoning, and bash commands in the terminal

Decision 3

Keep developers in control of AI actions.

Insight

Developers want assistance—not automation.

Tension

Autonomous AI → fast, but risky vs. user-controlled AI → slower, but reliable.

Decision

Kept developers in control—requiring review and confirmation before AI actions execute.

Trade-off

Reduced speed in favor of control and safety.

Developer confirmation prompt before writing a file
Developer confirmation prompt before executing a high-risk command

The system

How AI integrates into the developer workflow

Prompt → AI generates command → Review/Edit → Execute → Output → Iterate

Key Learnings + Impact

Impact

  • Reduced friction integrating AI into workflows
  • Increased usage of Cortex Code
  • Faster iteration cycles
  • More efficient debugging and querying
  • Increased confidence in AI-assisted workflows

Key learnings

  • AI needs guardrails
  • Transparency builds trust
  • Workflow integration > feature depth
  • Developers prioritize control over speed
  • Terminal constraints shape interaction patterns

AI is most effective when it integrates into existing workflows without breaking them.

To learn more go to Get started with Snowflake CoCo CLI