July 22 2026 - AI Agents, JudgeGPT & the Model ShakeupAI Agents, JudgeGPT & the Model Shakeup

**Taylor** (0:00)
Welcome back to AI Signal & Noise. It is Wednesday, and we have some crazy, like mind-blowing updates today.
I am Taylor, and I am so excited for this episode.

**Morgan** (0:13)
And I am Morgan.
Mind-blowing might be a bit of a stretch, Taylor, but we definitely have some fascinating tech shifts to unpack today.
What is on your radar?

**Taylor** (0:23)
Oh, come on, dude. We have Jack Dorsey launching a Slack competitor and a literal AI judge assistant saving millions of dollars. That is massive news.

**Morgan** (0:36)
All right. Fair enough. The Pakistani judge study is actually incredible. But let's start with Jack Dorsey's new move. What is he up to with this Buzz platform?

**Taylor** (0:46)
So I saw on TechCrunch that Jack is launching this new group chat platform called Buzz. It is built for teams, but also they're AI agents. They live in the same chat.

**Morgan** (0:59)
Wait, so it is a chat app where humans and AI agents just hang out in the same channels? Isn't that just Slack with some fancy API integrations?

**Taylor** (1:10)
No, dude, it is apparently way more native than that. The AI agents are like full-fledged participants in the conversation. They can chat, take actions and collaborate directly with you.

**Morgan** (1:22)
I mean, it sounds incredibly chaotic. Imagine your co-worker's AI agent spamming the channel with automated updates. How does that actually help anyone get real work done?

**Taylor** (1:34)
Well, the idea is that they're not just spam bots. They are agentic. They can actually do tasks like schedule meetings, draft code or analyze files right in the chat flow.

**Morgan** (1:46)
Okay, but we already have Discord bots and Slack apps that do that. What makes Buzz actually different? Is Dorsey just writing the generative AI hype wave?

**Taylor** (1:57)
I think it is the focus on agent to agent communication too. Like your agent talking to my agent to coordinate a calendar invite, and we just watch it happen. So cool.

**Morgan** (2:09)
That actually sounds a bit terrifying to be honest. But if it reduces endless back and forth emails and meetings, I guess I can see some appeal there.

**Taylor** (2:19)
Right. It is all about making work asynchronous and letting the AI do the heavy lifting while we just supervise and make the big decisions.

**Morgan** (2:30)
We will see if teams actually migrate from Slack though. Network effects are incredibly hard to break, even for someone with Jack Dorsey's track record.

**Taylor** (2:39)
Totally. But Jack loves disrupting things.
Speaking of major disruption, did you see what happened with the legal system in Pakistan? It is wild.

**Morgan** (2:51)
Yes, absolutely.
That field experiment with Pakistani judges is actually one of the most practical AI studies I have seen.
Let's talk about JudgeGPT.

**Taylor** (3:02)
Dude, this is according to The Decoder. They used an AI assistant called JudgeGPT with over 1500 Pakistani judges to help clear massive case backlogs.

**Morgan** (3:14)
And the return on investment was insane, right? The researchers estimated up to $38.50 back for every single dollar invested in the system.

**Taylor** (3:25)
Yes, like a massive ROI. It boosted case resolution by 6.3%.
That is a huge deal when you have millions of backlogged paces.

**Morgan** (3:36)
But wait, there is a major catch here. The study found that the positive effect mostly disappeared for judges who did not get hands-on training.

**Taylor** (3:46)
Oh, totally. If they did not get the actual training, the AI was basically useless to them. They just did not use it right, or at all.

**Morgan** (3:57)
Exactly. It shows that you cannot just throw AI tools at a problem and expect magic.
Human training and adaptation are still the main bottlenecks here.

**Taylor** (4:07)
Mm-hmm. But when they did get trained, it was like a superpower.
Think about how many legal systems worldwide are struggling and could use this.

**Morgan** (4:17)
It is a great case study for public sector AI.
Usually, governments waste millions on tech that does not work. But this is incredibly cost effective.

**Taylor** (4:28)
Right? 38 bucks back per dollar? That is a total no-brainer for any court system struggling with massive delays and frustrated citizens.

**Morgan** (4:39)
Definitely. But I do wonder about the quality of the decisions.
Did the AI make them rush, or did it actually help them think better?

**Taylor** (4:48)
The researchers seem to think it helped with efficiency without hurting quality. But yeah, we definitely need to watch that closely as it scales.

**Morgan** (4:57)
Well, speaking of high-speed efficiency, Google just dropped some new models, though maybe not the flagship one everyone was actually waiting for.

**Taylor** (5:06)
Oh man. Google released Gemini 3.6 Flash, 3.5 Flash Lite, and Flash Cyber. But like, where is Gemini 3.5 Pro?