Written by John Maeda Design in Tech Report 2026: From UX to AX — presented at SXSW 2026

All text below is the original, unedited work of John Maeda, open-sourced as part of the E-P-I-A-S × SAE Framework for AI upskilling product designers.

Moving from SAE L2 → SAE L3

This is the most important transition for product designers right now.

  • L2 = AI helps you build chunks
  • L3 = AI helps you run workflows

The shift is not “more AI,” it’s more structure, judgment, and repeatability.


SAE L2: 🚗🧠 Partially Automated

AI builds bounded chunks; you integrate and validate

At L2, designers are still very much “hands on the wheel,” but AI is doing meaningful production work.

What L2 looks like in practice

  • You generate screens, components, flows, or copy blocks
  • You manually stitch outputs together
  • QA is mostly visual + intuition
  • Success depends on how clearly you prompt

Typical L2 tools designers already use

  • App & UI generators: Bolt, Lovable, Framer, v0
  • Visual generators: Figma Make, Midjourney, Firefly
  • General LLMs: ChatGPT, Claude, Gemini (chat mode)

SAE L2 × E-P-I-A-S Growth Path

❶ Explorer → ❷ Practitioner (the first real win)

Goal: Stop “trying stuff.” Start getting repeatable results.

What to learn:

  • Write explicit instructions, not vibes
  • Separate input (requirements) from output (what AI returns)
  • Capture what works

Concrete upgrades:

  • Save prompts (even in Notes or Figma)
  • Define “done” for a generated component
  • Use the same prompt more than once

Signal you’ve leveled up:

“I can reliably generate this kind of component with predictable quality.”

❷ Practitioner → ❸ Integrator (where maturity shows)

Goal: Make AI output traceable and safe to integrate.

What to learn:

  • Break work into bounded chunks on purpose
  • Add manual QA checklists (accessibility, hierarchy, tone)
  • Document what the AI was asked vs what it produced

Concrete upgrades:

  • Prompt templates with sections: context / constraints / output
  • Explicit handoff notes: “This was AI-generated, reviewed for X, Y, Z”
  • Repeatable integration patterns (where AI output fits, where it doesn’t)

Signal you’ve leveled up:

“I can explain why this output is trustworthy.”

❸ Integrator → ❹ Architect (L2’s ceiling)

Goal: Other designers can reuse your L2 workflows.

What to learn:

  • Turn good prompts into reusable templates
  • Decide which chunks are worth automating
  • Design guardrails, not just prompts

Concrete upgrades:

  • Component-specific generators (forms, navs, empty states)
  • Prompt + QA bundles (“generate → check → refine”)
  • Shared libraries (Notion, Figma, repo, internal wiki)

Signal you’ve leveled up:

“People ask to use my AI workflows.”

❹ Architect → ❺ Steward (org-level L2)

Goal: Set standards for partial automation.

What to learn:

  • When automation helps vs hurts design quality
  • Risk areas (hallucinated UX patterns, accessibility misses)
  • Governance without killing speed

Concrete upgrades:

  • Team guidance: “Automate these chunks, not those”
  • Review standards for AI-generated UI
  • Mentorship on safe integration

The Transition: L2 → L3 (The Inflection Point)

This is where many designers stall.

What changes at L3:

  • You stop thinking in screens
  • You start thinking in runs

The key mental shift:

From “generate this” → “run this workflow”

SAE L3: 🚗😴 Guided Automation

AI runs multi-step workflows; you supervise and intervene

At L3, AI doesn’t just produce outputs — it follows process.

What L3 looks like in practice

  • You work inside an IDE or structured environment
  • You pass context, not just prompts
  • AI runs multiple steps in sequence
  • You add checkpoints and evaluations

Common L3 tools designers encounter

  • IDEs: VS Code, Cursor
  • Copilots & agents: GitHub Copilot, Claude in IDE
  • Workflow tools: LangChain, n8n, Foundry, simple scripts
  • Eval tools (lightweight at first): checklists, comparisons, diffs

SAE L3 × E-P-I-A-S Growth Path

❶ Explorer → ❷ Practitioner (most designers here today)

Goal: Make multi-step runs reliable.

What to learn:

  • Basic context engineering (rules, constraints, inputs)
  • Sequencing tasks (“first do X, then Y”)
  • Manual checkpoints

Concrete upgrades:

  • Move from chat to IDE-based workflows
  • Use system prompts / instruction blocks
  • Add explicit “stop and review” moments

Signal you’ve leveled up:

“My workflows don’t fall apart every other run.”

❷ Practitioner → ❸ Integrator (true L3 maturity)

Goal: Know when AI runs — and when humans step in.

What to learn:

  • Decision framing (“AI does this unless…”)
  • Lightweight evals (pass/fail, comparisons, constraints)
  • Exception handling

Concrete upgrades:

  • Simple eval checks (structure, length, criteria)
  • Clear ownership: AI generates, human approves
  • Documented failure modes

Signal you’ve leveled up:

“I trust this workflow until it triggers a known exception.”

❸ Integrator → ❹ Architect (L3 power users)

Goal: Others can run your workflows without you.

What to learn:

  • Modular context (inputs, rules, examples separated)
  • Reusable Skills or agent tasks
  • Shared eval patterns

Concrete upgrades:

  • Reusable workflow scripts
  • Context libraries (design rules, brand voice, constraints)
  • Tooling others can invoke safely

Signal you’ve leveled up:

“My system runs even when I’m not there.”

❹ Architect → ❺ Steward (org-level L3)

Goal: Make guided automation normal and safe.

What to learn:

  • Organizational risk and reliability
  • Training others on judgment, not tools
  • Maintaining shared infrastructure

Concrete upgrades:

  • Standards for IDE + AI usage
  • Mentorship on context engineering
  • Shared Skills, MCP tools, or workflow libraries

The Big Takeaway for Designers

  • L2 is about what AI makes
  • L3 is about how work flows

Most designers should:

  1. Go deep to Architect or Steward at L2
  2. Then move carefully into L3 Practitioner
  3. Build judgment before autonomy

Or said simply:

Don’t race to automation.
Design reliability first.