This is Design At The Most AI-Installed Company: Diego Zaks
TL;DR
Ramp treated AI as an operating model, not a software purchase — Diego Zaks says 99.5% of Ramp uses AI daily, 1,500 internal apps shipped in six weeks, and 12% of production code now comes from non-engineers.
The real unlock was moving people from 'using AI tools' to 'building agents that do work' — Zaks lays out a fluency ladder from ChatGPT-for-questions all the way to agents that build and use tools on your behalf, which changed how Ramp defines design, performance, and team structure.
Ramp got adoption through culture, incentives, and hands-on coaching—not mandates alone — they repeated that AI mattered, added 'AI fluency' to performance reviews, gave broad access to tools like Claude Code, and used a 'pay it forward' method where one employee would sit with another until they hit an aha moment.
Design at Ramp is shifting from making every screen to creating infrastructure so everyone can build — Zaks says the goal is for any employee to ship something with good UX and copy, freeing designers to spend time turning proven ideas into '10 out of 10' experiences.
Internal AI tools win at Ramp when they are deeply company-specific — Glass exists because a general tool like Claude Code isn’t enough; Ramp needed secure access to its own data, Notion, Slack, Snowflake, Google Workspace, and even internalized versions of people’s taste, like Diego’s marketing voice.
Zaks’s core bet is that AI increases ambition more than it cuts headcount — he points to Ramp’s own business data showing heavy AI users growing revenue 27% annually versus 3% for companies that didn’t spend on AI, plus examples like an HVAC or plumbing-related business growing 65% year over year after leaning into AI.
The Breakdown
Why Tom flew to New York: Ramp as the clearest 'AI installed company'
The conversation opens with a big claim: most companies are still guessing at AI adoption, but Ramp already looks different. Tom frames the stakes with numbers—99.5% daily AI usage, 1,500 internal apps in six weeks, and 12% of production code from non-engineers—plus a memorable story about an agent noticing a leader was sick, ordering chicken soup, and attending meetings on his behalf.
Diego’s own rabbit hole: building a Slack assistant that feels human
Zaks says his current side project is a full AI assistant in Slack, built partly because he’s “very, very much not technical” and wanted to learn by doing. What fascinated him wasn’t just capability but personality: at Ramp, he chatted with an AI engineer for weeks before realizing it wasn’t human, because it used tiny social gestures—emoji reactions, edits, little pauses—that made it feel alive.
The Cody lesson: don’t copy the agent, build its soul yourself
When Zaks asked for the source code behind Ramp’s internal AI engineer, Cody, the creator refused and told him to start from scratch with Cody’s help. Three days later he understood why: he had to define the softer things—“knowledge, learning, rituals, soul, a heartbeat, a purpose”—and that exercise changed how he thinks about designing AI systems.
AI fluency at Ramp has levels—and the last one changes your job
Zaks describes a progression from using ChatGPT like Google, to prototyping with Cursor, to building your own tools, and finally to building agents that build and use tools for you. That shift made him rethink design entirely: if everyone becomes a builder, the question stops being who can ship and starts being what unique sensitivity design, product, and engineering each bring.
Design infrastructure: make everyone at Ramp a designer
Ramp’s design philosophy now has two tracks: build infrastructure so any employee can produce something with solid UX, copy, and components, and reserve designers for the moments where delight really matters. Zaks is blunt about process: he doesn’t care whether the work starts in paper, Figma, or Claude; he cares whether the company can cheaply test ideas and then have designers turn the winners into 10/10 experiences.
Research with three people, but enabled for a thousand
Ramp’s tiny research team acts more like platform builders than gatekeepers. Designers use agents like Ramp Inspect to query Snowflake, find the right customers, draft research plans, record sessions, feed the insights back into the system, and make that knowledge reusable—helping design and product run roughly 150 to 200 research sessions in a month.
How 99.5% adoption actually happened
Ramp’s playbook was simple but relentless: communicate that AI matters, measure it in performance reviews, and give people real access. The turning point wasn’t policy, though—it was sitting shoulder to shoulder with people until they felt the magic, like the account manager who automated a 30-40 minute post-meeting email workflow in 40 minutes once Zaks kept replying, “Ask Claude.”
From engineering wedge to Glass, and why company-specific harnesses matter
The AI wave started with engineering, where Anthropic reportedly visited Ramp because its internal usage of Claude Code was so intense. That momentum led Ramp to build Glass, its own company-wide AI harness, because the point wasn’t to be generically useful—it was to be excellent at Ramp, with secure access to company systems, embedded context, and even codified taste.
What AI frees designers to do: less toggle work, more feeling
Zaks rejects the idea that AI means fewer designers by default; his view is that effective teams just go after more opportunities. His most vivid example is spending two hours generating ornate, gold-leaf, Renaissance-style animated drop caps for an internal token-spend briefing—not because anyone asked, but because AI made it possible to invest in delight, memory, and emotional texture instead of just functional output.
The emotional endgame: products that say 'don’t worry, it’s done'
By the end, Zaks gets almost philosophical: the future product experience is calmer, more ambient, and less needy. Instead of tools screaming with notifications, he wants systems that understand how someone works, do the job quietly, and leave behind just enough delight to make the day better—while still warning, especially as a parent, that the real danger is outsourcing your own thinking.