AI and the Future of Work
Six levels of being an "AI-native" organization
From Ann Miura-Ko:
"Questions around what is truly AI-native reminds me of the debates we used to have about the levels of autonomy in autonomous vehicles (AVs). For years, everyone in AV was chasing Level 5 self-driving. The levels mattered because they forced precision. Cruise control was not autonomy. Lane keeping was not autonomy. Driver assistance was not the same thing as self-driving."
Miura-Ko lays out six levels for how deeply an organization has rebuilt itself around AI, using four questions: what can AI see, what can it do, who can extend the system, and how has the organization actually changed?
The levels:
L0: AI as theater. The CEO gives the speech. Nothing in the org actually changes.
L1: Personal productivity. People use ChatGPT solo. If your best AI user quits, the workflow leaves with them.
L2: Team workflow. Each function builds its own private AI stack. The stacks don't talk.
L3: Organizational infrastructure. Agents act across systems of record. Non-engineers author shareable skills.
L4: Compounding operating system. The system learns from its own runs. Non-engineers ship production internal tools.
L5: Virtually self-driving. The system notices, decides, acts, and updates itself. Humans govern strategy and taste. (Miura-Ko admits L5 doesn't exist yet.)
The reality is that most companies are only at the beginning of truly using AI, mostly for personal productivity or team tasks. Becoming AI-native means deeply embedding AI into how the whole organization works.
Ann Miura-Ko | Everyone wants to be AI-pilled. Most companies are still Level 1
☕ Medium Read (7 minutes)
Related from Ethan Mollick: “Organizations are already superhuman intelligences. The University of Pennsylvania, Walmart, or whatever is far more capable than any human at a wide range of tasks. That is why the focus on AIs as individual productivity tools hits a natural limit, many benefits of AI depend on integration with firms, and the people who populate them, in ways that strengthen both.”
Designing with AI
Inside the messy reality of design tooling at Vercel
Hannah Hearth, on the Vercel product design team, hosted a tooling show-and-tell and wrote up what her colleagues are actually using. Her findings:
Even on her small team, everyone uses completely different tools right now.
AI usage spans the full range, from designers who haven't gone a day without it in 6 months to skeptics who still find it mostly useless creatively.
AI tools for design workflows are clunky and lag behind those for engineering.
The takeaway: There is no consensus, and you're not the only one still figuring this out. A few specifics:
One team member runs Conductor (a Mac app for orchestrating several Claude Code agents in parallel) with multiple tabs in one project, one for UX flows, one for state and data, one for paper-cut fixes, so he stays in flow while his agents work.
With production as "the source of truth," the team needs to get production styles back on the canvas. They've had success with the Paper browser extension, which copies a page's structure and styles and brings them into the Paper design canvas.
Figma still plays a role, but that role varies by designer: some prefer to explore in Figma, while others find it dramatically slower than AI tools.
The whole post is worth a read for a real-world perspective.
Hannah Hearth | Tools the Vercel product design team actually uses
⚡ Quick Read (5 minutes)
Tools and Resources
Couch to 5K, but for Claude Code
Hilary Gridley's Couch to 5K for AI is a free 30-day program built around 10 minutes of practice a day. The program walks you from chatting with AI to actually building things in Claude Code, with a daily prompt you can easily finish before your workday starts. The approach borrows from fitness programs: small wins, daily reps, and momentum that compounds.
If you've been struggling to find time to start building with Claude Code, this program is for you!
Couch to 5K for AI
💡 Bookmark for Reference
AI Agents
Clarie Vo's agent-vocabulary-in-a-Tweet
Claire Vo dropped a quick glossary covering the working vocabulary of AI agents: message, agent, sandbox, subagent, tool, hook, connector, MCP, skill, plugin, harness, agent SDK. Each term gets a one-liner. She concludes: "what did I miss and also what is wrong with us." A good question for anyone trying to keep up with this vocabulary!
And then of course come the replies: people chiming in with clarifications, additions, and "well, actually" footnotes.
X (@clairevo) | Quick agent terminology glossary
⚡ Quick Read (1 minute)
That’s it for this week.
Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋
All the best, Heidi
