How Rippling learned to work differently

How curiosity turned into Rippling’s operating system for AI

Every transformation looks obvious in hindsight.

Nine months ago, one engineer opened a Cursor prompt window.

Not a committee, not a strategy offsite, just curiosity and an idea worth testing. The result? A few hours saved.

Small on paper, but that single experiment lit the fuse for how Rippling would learn to work differently.

AI is changing how we work, not because it automates tasks but because it changes how we think.

At Rippling, we’ve treated that shift as a design challenge: start small, share what works, and scale curiosity into leverage.

This is the story of how that unfolded: how a handful of bottom-up experiments in recruiting, product, legal, engineering, and design evolved into a new way of building, deciding, and learning together.

1. It started with curiosity

When I joined Rippling, AI wasn’t yet a formal initiative. Product teams were already building AI-powered features for customers, but internally, adoption of AI tooling was emerging bottom-up.

Across the company, people were quietly experimenting:

• An engineer using ChatGPT to debug a tricky issue.

• A recruiter summarizing intake notes in NotebookLM.

• A PM running tone checks on product copy.

Each spark was isolated but promising.

At Rippling, leadership starts with going and seeing, getting close to the work. Albert Strasheim and I realized that adoption wouldn’t come from a strategy deck. It would come from first-hand experience and visible permission.

So we decided to set a simple stance. We gathered a small group of early adopters, wrote down the principles on a single page, and published them company-wide:

• Use AI wherever it helps you work faster or smarter.

• Treat AI like a new hire: give it context, define success, and review its work.

• Test everything before you rely on it.

• You are accountable for the output.

It wasn’t a manifesto. It was permission, a license to experiment safely.

Once we said that out loud, usage exploded. Within days, teams were sharing workflows, results, and new ideas.

2. Curiosity built community

The next question was simple: how do we learn from one another?

Step one was to find the people already ahead.

We analyzed internal usage data across tools like Cursor and Gemini to identify early adopters.

Engineers, recruiters, and designers who were already building their own AI workflows became our first AI champions.

Instead of slides or frameworks, we asked them to show their work live. That first 45-minute session became SPARK: Spotlight on AI at Rippling.

An engineer in India demoed how she used Whisper to talk to Cursor. Another showed how he built a Go service even though he only knew Python. A designer walked through an AI-assisted copy review that streamlined dozens of micro-edits.

The energy shifted instantly. People stopped watching AI and started working with it.

That’s what leadership looks like here: deciding fast, challenging ideas in the open, and learning in real time.

Within days, new Slack channels appeared. Engineers began sharing prompt packs. Recruiters shared NotebookLM templates. Designers iterated on GPT prompts for tone reviews.

SPARK turned curiosity into community. It was never a program. It was a learning loop, a belief built through doing.

3. When learning scaled 

Once belief spreads, systems catch up.

Teams began formalizing what they had been doing ad hoc.

Managers launched weekly AI coding hours.

Others ran 45-minute sessions where everyone brought one piece of current work they disliked, delegated it to AI, and shared what worked.

Across Rippling, scattered wins turned into shared systems.

• Recruiting team: built NotebookLM notebooks that auto-summarized interviews. Prep time dropped from hours to minutes, and new recruiters started from a shared context instead of blank pages.

• Legal team: created prompt libraries for contract review. Reviewing terms that once took days now happens in a single working session.

• Product and Design: collaborated on a Product Copy GPT that reviews language across every screen, maintaining consistency and tone at scale.

• Engineering: combined Cursor, Claude Code, and Gemini to debug incidents, summarize PR threads, and generate tests in minutes. The mix became a live example of “minutes, not hours” in action.

Everywhere, people stepped outside their lanes to make the system better.

That is the Rippling version of going to Western Union — doing what needs to be done.

This was the moment Rippling stopped trying AI and started transforming through it.

The shift from isolated tools to a new operating rhythm.

4. Craft, pilots, and the build + buy flywheel

More output isn’t the goal. Better output is.

We quickly learned that AI slop is real.

At Rippling, taste became the differentiator.

Generate ten options, pressure-test them, iterate a hundred times if needed.

That principle now shows up everywhere, from marketing copy to product reviews.

In engineering, BugBot now reviews more than 9,000 pull requests a week, flagging small but critical issues. Over half of those findings are fixed automatically, and thousands of reviewer minutes returned to deep work.

The goal isn’t speed alone. It’s raising the floor so every engineer can operate at the level of our best ones.

That is what it means to push the limits of possible: higher craft at higher velocity.

Pilots Over Programs

Every major shift here begins as a pilot, small, fast, and measurable. We don’t launch programs. We run experiments.

AI changed the economics of building and buying software.

In the old world, building custom meant months of engineering work and high opportunity cost.

Buying tools off the shelf was faster but rigid, hard to customize, and rarely fit our exact context.

Now we can pilot dozens of AI tools, treat them as experiments, and learn what fits our environment.

If a tool delivers outcomes, we double down. If it doesn’t, we move on.

What makes this work at Rippling

• Run rapid pilots. We evaluate tools in weeks, gather real data, and make fast calls on what stays or goes.

• Co-design with startups. We partner early. With Astral (the team behind uv for Python), we optimized model performance at enterprise scale. We collaborated with Cursor to evolve BugBot metrics and shared data with OpenAI to study Codex behavior in large environments. This isn’t vendor management. It’s co-creation.

• Set clear end criteria. Every pilot has a defined success metric. Some work, some don’t, and that’s fine. Each experiment teaches us what a tool can do versus what it actually does in our context.

These pilots don’t just validate tools. They shape our products.

Every working internal pattern becomes a candidate for what we ship to customers, tested, refined, and proven in real workflows.

That discipline: fast pilots, high taste, real feedback, and deep partnerships is what makes experimentation at Rippling sustainable.

Responsible innovation

We are intentional about where and how experiments happen.

Every AI pilot, no matter how promising, goes through a joint legal, security, and procurement framework before it touches a critical workflow.

That process ensures responsible innovation: teams can move fast without creating hidden risk, duplicate spend, or vendor creep.

It keeps us creative and safe at the same time.

We learn without losing control.

It’s how we scale frugally, decide fast, and stay right a lot.

The flywheel in motion

The result is measurable.

We’ve grown from 300 engineers using Cursor to more than 1,200+ daily active users in just a few months with 50%+ AI-flagged findings closed. PMs and designers are now part of that loop. 

We use AI to make Rippling better, then build AI that makes our customers better.

5. How AI shows up in the product

The same loops that changed how we work are now shaping what we build. AI isn’t an add-on at Rippling. It’s becoming part of how the product itself thinks.

1. Unstructured information processing

Rippling Recorder transcribes and summarizes interviews, giving teams structured insights instantly. The same architecture powers expense categorization, policy lookup, and document parsing across HR and IT. Teams that once spent hours organizing inputs now focus on decisions, not data cleanup.

2. Talent signals as add-on value

By connecting data across payroll, recruiting, and performance, we surface actionable signals from hiring velocity to attrition risk, helping leaders make faster, data-backed decisions. Every new signal compounds the context of the system, turning workflows into intelligence.

3. Agents that help users day to day

Built on the Employee Graph, Rippling’s agents understand context across systems. They can answer employee questions, build reports in plain English, or trigger workflows like creating a pay run or scheduling interviews. Users no longer have to learn the system. The system learns them.

These are not features stacked on top of a product. They show how the product itself is learning to think and act.

At Rippling, craft means building AI that feels invisible, useful without being loud, and powerful without being complex.

6. The Rippling flywheel

Our hypothesis is simple.

People who use AI start to think differently, and that changes the products they build.

That is the loop powering Rippling today.

Every engineer who uses AI to code faster expects the same leverage in our internal tools.

Every manager who runs an AI pilot designs new team rituals around it.

Every product team that sees those patterns builds them directly into Rippling.

That is our flywheel:

Use AI → Think with AI → Build AI-native products → Enable the next user.

Each turn compounds our advantage through taste, judgment, and firsthand experience. Rippling continues to evolve as a company that learns faster than anyone else because it learns by doing.

We also learned something else: Speed without discipline eventually slows you down. That is why we pair experimentation with guardrails.

Every AI pilot is reviewed jointly by our legal, security, and procurement teams before it scales. We end pilots that do not hold up, no matter how exciting they look on paper.

Frugality and governance are features, not constraints. They are the reason our innovation compounds safely rather than spinning out of control.

Transformation is not a milestone. It is a muscle.

Rippling’s AI journey did not start with a strategy document or a task force. It started with individuals taking agency, trying things, sharing what worked, and holding a high bar for quality.

We learned to lead by example, to go and see, to decide fast, to push what is possible, and to build teams that win together.

AI has not changed that. It has simply given us more leverage to do it better.

If thirty minutes together can save your team a hundred hours, take it.

That is how change happens here, one spark at a time.