Happy Tuesday.

I’m a big AI subscription guy.

But buying AI licenses is not the same as installing AI into your work, and that became clear to me when I flew to NYC to talk to Diego Zaks.

He runs design at Ramp, a company that received a visit from Anthropic because they were using Claude Code more than Anthropic was.

This is what design looks like at a 1,500-person company using AI at a scale you haven’t seen.

– Tommy (@designertom)

Framer just shipped a major CMS redesign and called it "only the beginning."

Framer's pitch is design-speed content infrastructure: inline editing, structured collections, SEO, and collaboration in the same place the page is already being designed. No second system to explain. No after-the-fact wiring. The page and the content model ship together.

If your CMS still feels like the meeting after the meeting, this is the other path.

Try Framer →

Diego said it plainly: "I don't need AI tools to do my job. I need AI to do the work."

That line is useful because most AI advice for designers is stuck one rung too low.

Anthropic's AI Fluency Index studied 9,830 Claude conversations and named 24 behaviors that make people better at working with AI. The floor is rising fast.

And according to Diego, AI fluency has levels.

Many designers think they're on level three or four because they shipped a prototype with Cursor or made a Lovable app. But it might be helpful to consider the levels:

You ask questions, summarize, research, and replace a search bar with a chat box.

You use Cursor, Claude, Lovable, or v0 to make a prototype, draft, screen, script, or throwaway internal tool.

This is where a lot of designers are now. Old friction. If the next project needs the same setup, the same prompt chain, the same five tabs, and the same manual glue, the AI helped you make an output. It didn’t change how you work.

You build your own tools: skills, agents, and little command centers you can call from Slack or the terminal.

The work gets faster every week because the scaffolding compounds. You stop asking the model to help with one task, because you’re able to build an entire system around the task.

This is where the taste work gets interesting, because you're judging the machinery that produces large batches of probabilistic attempts.

What blocks level four: the work still flows through you. You are still the bottleneck. And that’s not to say this is good or bad. It’s just true.

Agents build and use tools to do the work. You define outcomes, taste, constraints, and review loops. The work runs without you for longer durations of time.

This is still uncommon because most companies don't have the harness for level four to exist inside of it.

The reason most designers are stuck at level two is structure.

I wrote about this in the next gap in design work: designers are too employed to climb. No experiment wiggle room or token budget.

A Cursor prototype is level two. A skill that runs Cursor for you is level three. A Slack agent that runs the skill while you're in a meeting is level four.

Anthropic's data says only 30% of users set collaboration terms upfront. When the output looks finished, people stop evaluating it.

Ramp is the cleanest field report I've seen.

Eric Glyman shared that Ramp hit 99.5% daily AI usage, shipped 1,500 internal apps in six weeks, had non-engineers write 12% of production code, and created agents that order chicken soup and show up to meetings, not as a note-takers, but as stand-ins for sick employees (unprompted).

Diego also built his own AI assistant in Slack despite being, in his words, "very, very much not technical," because he’s trying to understand how to get people to stop using the current Ramp product.

Ramp Labs keeps publishing the receipts too: the gold-leaf AI token spend briefing, and a practical guide on agents users can trust.

They’re one of the few companies publicly sharing how they’re installing AI at scale.

These are tools I actually use, so I asked them to sponsor the newsletter. They said yes. The best way to support us is to check them out 👇

That's it for today.

Friday I'm going deeper on the company shape that makes level four possible.

Reply with the level you're actually stuck on.

See you then,

Tommy