How Amp Ships 50 Times a Day With AI Agents

SPEAKER_1 (0:00)

This episode is brought to you by Paypal. You know how a mom's bag has everything? Sunscreen, snacks, a stapler?

The new Paypal app is like that, but for your money. Shop, pay, manage your account, and earn rewards all in one place. And with purchase protection on eligible items, biometric security, and pass keys, you're protected at every step. Download the new Paypal app to get started. See paypal.com/protectionterms.

SPEAKER_2 (0:29)

When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen, and helps reach people with the right skills, certifications, and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsored job credit at indeed.com/podcast. That's indeed.com/podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs.

Dario Hamidi (0:59)

You can get double digit percentage, if not triple digit percentage gains in productivity, if you get rid of poll requests in your organization.

Patrick Akil (1:07)

This is Dario Hamidi, AI engineer over at AMP, and he walks us through some of the biggest problems he solved at AMP specifically. Shared agent state, remote agent execution, and how agents handle permissions.

Dario Hamidi (1:19)

Part of your job is to create an environment in which you can trust the agents. How about we don't juggle knives? What parts belong into the model layer, and what parts belong into everything that's deterministic and around it? And that's part of the agentic engineering skill set.

Patrick Akil (1:33)

They deploy the production about 50 times a day, and they do so without any poll requests in between. Is it time to kill the poll request?

Dario Hamidi (1:41)

Beyond Coding One of the key enablers for our productivity to even be in a spot where we can get so many bug reports because we ship so much, is actually not doing feature branches, not doing poll requests. And I would even go as far as saying, even if you're not using AI today, you can get double digit percentage, if not triple digit percentage gains in productivity if you get rid of poll requests in your organization.

Patrick Akil (2:13)

So just merge? Yeah. Okay.

How do I get there? To be able to confidently just merge?

Dario Hamidi (2:20)

How big is your team?

Patrick Akil (2:21)

Let's say my team is eight people, eight engineers.

Dario Hamidi (2:24)

If you told them the following tomorrow, that every poll request will get auto-closed, and to get something out, you have to commit and push to main. What do you think will change in people?

Patrick Akil (2:47)

People are going to be double checking all of their work themselves first and foremost. Likely going to do agentic verification. I don't know if they're going to ask a colleague in a session to verify the work, but in the end, I think the outcome will be that they have to be more confident pushing their change because it goes to main directly.

Dario Hamidi (3:03)

Yeah. What ways of increasing confidence beyond more and more verification can you think of?

Patrick Akil (3:11)

For me, all my thoughts immediately go into verification. So if I do agentic work nowadays, one of the things because I'm lazy and I don't want to do many pull requests, is I ask my agent to evidence work. So if we're working on a functionality, can you show me screenshots? Can you show me video? Can you show me this thing working? So that's more of the evidencing, but that's still in line with verification.

What I've seen agents being really good at is if there's a lot of deterministic tooling around how we want software to look, or how we want software to behave.

So an agent can work within those boundaries and be more effective. So I don't really have to do any nitty gritty details on how code looks, because it's structured through deterministic.

Dario Hamidi (3:48)

It's definitely like verification is definitely a big part. What I was trying to hint at is that the psychology of making changes also changes towards making smaller changes, because smaller changes are less risky. So if you feel the desire to open up a pull request, that's a signal to you that something is off.

That something is off, and you need to think about why you want to open a pull request. Obviously, something motivated me to open a pull request. Usually, it's fear. You're afraid that if you push your changes to main, they will break something. And then how can you get rid of that fear? Often, the fear comes from the changes being too big.