Matthew Braga

speaker
322 appearances 6 recordings 1 series first heard Jun 2026 last heard 11 Aug

Matthew Braga’s voice in public audio — every appearance, attributed to the second.

Trend

recordings per month · last 12 months
3 · Jun OctJan 26AprJulnow

Recordings per month over the last 12 months — 6 in all, peaking in Jun 2026 with 3.

Appearances

newest first · ▶ plays the moment
So there's the chat experience that the user sees, and then you have the AI agents that power it, that fetch and reason and sift through your content to find that one elusive nugget of knowledge you need.
And that's where Jongman comes in.
You've been at Dropbox for a long time now, 12 years, in fact.
What has kept you here?
Which in other words means you're not necessarily going to get exactly the same answer or response each time, right?
How do you ensure that an agent actually does what the user expects it to?
Is the idea that you keep refining that prompt until it gets closer and closer to what a human would rate the response or the output?
That's right.
Okay.
And how are we at that?
Okay.
So there is layers of automation on automation basically happening here.
That's right.
So by going through this process, this is how you're able to measure whether the outputs you're getting are getting better or worse over time?
As the models that we're using continue to get better, does that make the quality that we get of that evaluation better as well?
Of course, because the models aren't deterministic, it means the results are going to be a little bit different, like the delivery may vary.
And so through the evals, you're looking more for consistency of the response rather than the response being exactly the same every time.
Speaking of fun challenges,
How do you strike that balance between letting the agents do things autonomously on a user's behalf, but not giving it so much agency that it can, you know, get into some trouble, you know, like deleting all your documents or sending a rude email?
When we are talking about all of the infrastructure that we are building to support agents, allow us to run agents, are you able to break those systems down or like just kind of help explain sort of where does everything fit?
Showing 201–220 of 322 · page 11 of 17 ← Previous Next →