What it looks like when everyone builds with AI

At Samsara, building with AI has become part of everyone's job. The work itself is changing as designers, product managers, engineers, and leaders find new ways to build new products and solve problems for our customers.

Across the board, we're “all in” on AI, giving everyone the tools to experiment every day. People share what they're learning, build on each other's ideas, and help new ways of working spread quickly across teams. Leaders are in the details too, building alongside their teams, learning with them, and helping shape how we work.

In the latest episode of Object Stats, Chief Technology Officer John Bicket and Chief Product Officer Johan Land dig into the AI shift and what that actually looks like across their teams: the new ways of working, the unexpected challenges, and what the next phase of building looks like as the tools keep evolving.

Watch Video

🎧 Listen to the full episode now and subscribe to Object Stats wherever you get your podcasts: Apple, Spotify, Amazon Music.

Here's what we learned:

1. When everyone becomes the builder

Designers are no longer handing off to engineers; they're shipping code themselves. All of Samsara's designers are set up with Cursor, many have code in production, and the prototypes they're building are fully clickable products inside the Samsara dashboard. Getting something like that in front of a customer used to take weeks. Now it takes hours.

On the PM side, the shift is less about building and more about depth. Samsara runs roughly 1,000 Gong calls a day. The expectation now is that PMs have effectively reviewed all of them, not by reading, but by running agents across the data to surface what actually matters.

The engineering role is expanding just as quickly. Engineers are managing teams of agents, exploring more solutions in parallel, and spending more time solving problems than writing code.

2. From writing code to managing agents

The new engineering rhythm looks nothing like it did a year ago:

• Three to five projects running in parallel at any time

• PRs as a form of communication, not just a deliverable

• Tech debt tackled in massive parallel, months of work done in days

Engineers hone an agent to make a change, verify it works on a few packages, then run it across the entire codebase. And rather than writing code, most of the work is now exploring the problem space: two weeks of back-and-forth with agents, then a day or two to execute once the right approach is clear.

3. The drift problem: why sycophancy is more dangerous than it sounds

LLMs are trained to give the answer the human preferred, which means they'll confirm your ideas, tell you you're making progress, and steer you toward what you want to hear. Johan calls it drift: you iterate in a direction that feels right but hasn't been tested against reality.

The fix is building in truthful checkpoints along the way. A test, a different agent reviewing the work, or a measurable output that proves something actually worked. The exploration phase can be fast and loose. The conclusion needs to be human-reviewed.

4. What comes after agents

The tools everyone is using today are just the starting point. The next phase is companies building their own agent stacks on top of them: custom context, institutional knowledge, and skills tuned to how their systems work. Engineers are already thinking about how to divide up a project across agents and how to review work at that scale. That's a fundamentally different job than it was a year ago.

At Samsara, we invest heavily in AI and push the whole company to move with it, not just engineers. The expectation is that everyone pushes what's possible. We build for some of the most demanding operations in the world, and the way we work reflects that.

Come build with us.