AI Belts Guide [v1.0]

<aside> 🔄

This doc defines AI Coding Developer Belts — a progression guideline to help engineers grow their AI-assisted development skills. The goal is to empower every Notino to delegate more confidently to AI, ship faster, and build shared expertise.

This is part of the AI Dev Buddy DB program, which helps us set up mentorship for Notinos to pair and improve their dev workflows.

</aside>

Generally you should spiritually meet ~90% of the description to meet a level and the levels before it. Use your best judgement, this is all self-reported!

<aside> ⬜

White — Starter

Beginning to use AI suggestions and agents. Learning when to accept or reject completions.

• IDE inline completions enabled (e.g., Cursor Tab)
• Accepts suggestions for boilerplate, syntax, small snippets
• Experimenting with asking agents to do small, contained tasks

You know you're here when: You've accepted many AI inline suggestions and are new to using agentic development.

</aside>

<aside> 🟨

Yellow — Novice

Actively delegates small tasks to agents. One task, one agent, full attention.

• Asks agents to write functions, tests, or draft PRs
• Asks AI questions to understand outputs and navigate the codebase
• Growth mindset: Recognizes prompting is a skill; has read resources and applied tips with noticeable improvement

You know you're here when: You've shipped multiple agent-assisted PRs and can point to a prompting technique that noticeably improved your results.

</aside>

<aside> 🟦

Blue — Intermediate

Delegates real coding tasks end-to-end. AI writes first drafts; you review and refine.

• Runs agents with auto-approve for larger tasks
• Agent produces full first-draft PRs; you plan, review, refine, and merge
• Delegates substantive feature work, not just boilerplate
• Context curator: Writes detailed prompts and provides files, screenshots, docs, and uses Skills to improve output
• Verification loops: Prompts AI to write and run tests before handoff

You know you're here when: You spend meaningful time planning and crafting prompts, regularly leave an agent to complete a long task end-to-end, and return to review.

</aside>

<aside> 🟫

Brown — Advanced [Eng goal]

Delegates work confidently with self-verifying loops, sometimes on many parallel tasks if applicable.

• Uses AI effectively across varied scenarios: debugging, prototyping, incidents, refactors, most daily coding work
• Structures prompts so AI self-verifies and chains multiple steps; comfortable walking away for extended periods (>10 mins)
• Spends more time planning and reviewing than writing code
• Comfortable running multiple agents concurrently on separate workstreams
• Improves shared context: Edits skills, rules, and docs to benefit the whole team
• Mentors others: Has shared prompts, techniques, or lessons with at least one teammate

You know you're here when: You can orchestrate multiple agents and walk away confidently, and have improved the shared context or mentored others.

</aside>

<aside> ⬛

Black — Expert

Elevates the system for others. Pushes Notion's AI devtools culture forward.

• Strong intuition for how to scope and delegate any problem to AI, helps teach this to others
• Orchestrates many concurrent agents on long-running, self-verifying workflows
• Builds the harness: Contributes shared prompts, rules, skills, or tools that others adopt
• Scales knowledge: Creates guides, leads workshops, or builds training that reaches many
• Explores the frontier: Tries new models, techniques, or workflows and shares findings early

You know you're here when: Many Notinos use prompts, tools, or guides you created. You try new AI workflows months before they're widely adopted.

</aside>

Commentary

<aside> 💡

Why this framing?

The core progression encourages moving from single-threaded to multi-threaded progression, deeper levels of delegation, and becoming more empowered with context and verification. It’s not perfect, but we wanted to highlight these key skills progressing as you move to higher belts.

Evolving as we go!

These definitions will evolve as tooling improves and people become more comfortable with AI. It's okay to recalibrate as the definitions are refined. The goal is to build shared understanding and cultivate growth mindsets, not to judge performance.

Note that this is forward looking guidance for this half (H1 2026), accounting for the fact that models and tooling will improve rapidly. We may re-calibrate in the future.

</aside>

Agent Delegation Levels

This is a diagram representing 8 different agent delegation levels from Steve Yegge’s *The Future of Coding Agents:*

<aside> 🎯

Your goal should be to learn and build comfort with higher delegation levels and use them when appropriate, not to spend all your time at higher levels if it doesn’t make sense.

Examples:

• One Yellow belt engineer might be mostly operating in Figure 1-4 (80%) and experimenting with Figure 5 (20%).
• One Brown belt engineer might be operating at Figure 6 (50%), Figure 5 (30%), Figure 1-3 (20%). They might use Figure 6 to work on multiple long running tasks, switch to Figure 5 to focus on iterating on one feature, then swap to Figure 1-3 for cleaning up and addressing PR comments.
• One Black belt engineer might be experimenting with Figure 7 and 8 on side projects, exploring ideas like Gas Town, but operate like the Brown Belt engineer above otherwise.

Notes:

• Figure 7 and 8 are still experimental and frontier. Operating at Figure 7+ may be unproductive for the vast majority of people at this time working in notion-next.
• You may also choose to run worktree agents (Figure 5+) in Cursor or some other IDE as well. Consider this spiritually as Figure 5+. 😛 </aside>