GopherCon LATAM 2026: AI Orchestration and 'Green Tea' GC Performance

Host: Alex Chan Guest: Mateo Silva Host: Hey everyone, welcome back to Allur, your weekly deep dive into the world of PHP, Laravel, Go, and everything mobile. I’m your host, Alex Chan. Today, we are talking about something that has been buzzing in my newsfeed all week—and if you’re a Gopher, you probably already know what I’m talking about. GopherCon LATAM 2026 just wrapped up in the absolutely stunning coastal city of Florianópolis, Brazil. Now, usually, when we talk about Go conferences, we’re looking for small iterations on the standard library or maybe some niche tooling. But 2026 feels... different. It feels like a "coming of age" moment for the language. We’re seeing Go move from being just the "language of the cloud" to becoming the heavy lifter for AI agent orchestration. Plus, there’s this new Garbage Collector version in Go 1.26 called "Green Tea" that is apparently breaking all the old performance benchmarks. It’s a lot to unpack, especially how the Latin American dev community—particularly the fintech giants in Brazil—are now the ones actually driving the Go core forward. To help me make sense of all this, I’ve got a very special guest joining us today. Host: I’m thrilled to introduce Mateo Silva. Mateo is a Lead Principal Engineer at one of the largest fintech platforms in São Paulo and was actually on the ground in Florianópolis this week. He’s been a contributor to the Go runtime and has a front-row seat to how these new 1.26 features are being used in high-stakes financial systems. Mateo, welcome to Allur! Guest: Hey Alex! Thank you so much for having me. It’s great to be here. Honestly, I’m still a bit jet-lagged from the flight back from Floripa, but the energy at the conference was so high, I think I’m still running on that adrenaline... and maybe a bit too much Brazilian coffee. Host: Oh, I bet! I’ve seen the photos—the beach, the tech setups—it looked incredible. But let’s dive into the meat of it. The big talk of the conference was Go 1.26. We’ve been hearing about this "Green Tea" Garbage Collector. First of all, why "Green Tea"? And what makes it so different from the GC we’ve been using for years? Guest: (Laughs) Right? The name "Green Tea" actually started as an internal joke among some of the contributors because it was meant to be "refreshing" and "lightweight" compared to the older, more aggressive collectors. But technically, it’s a huge shift. For a long time, Go used a concurrent mark-and-sweep GC. It was good, but when you hit these massive heap sizes in high-frequency trading or complex AI systems, you’d get these p99 latency spikes that were just... painful. Host: Right, those "Stop-The-World" moments where everything just hangs for a millisecond? Guest: Exactly. And in fintech, a millisecond is an eternity. "Green Tea" introduces a much more refined generational approach and a new way of handling the write-barrier. At the conference, the Go team showed data where these latency spikes were reduced by up to 40%. But the "aha moment" for me was the stack allocation. The compiler in 1.26 is just so much smarter about escape analysis. It keeps way more objects on the stack instead of letting them "escape" to the heap. If it’s on the stack, the GC doesn’t even have to look at it. We’re seeing a 20% increase in requests-per-second on our gRPC services back at the office without changing a single line of our code. Host: Wait, 20% for free? Just by upgrading the runtime? Guest: Literally. You just recompile with 1.26 and—boom—the "Green Tea" GC and the new stack allocation just make the binary leaner. It’s the closest thing to a "free lunch" I’ve seen in my career. Host: That’s wild. Actually, speaking of things that usually aren’t free—AI. There was a lot of talk in Florianópolis about Go displacing Python for AI orchestration. Now, usually, Python is the king of AI. Why is Go suddenly winning the "agentic workflow" battle? Guest: This was probably the most heated discussion in the hallways! Look, Python is still king for training the models and doing the heavy data science. No one is arguing that. But once you have the model and you need to build "agents"—these systems that have to reason, call three different APIs, wait for a response, and then make a decision—Python starts to struggle with the concurrency. Host: Because of the Global Interpreter Lock and all that? Guest: Exactly. But Go? Go was *born* for this. When you have an agentic workflow, you’re basically managing hundreds of concurrent "thought processes." In Go, these are just goroutines. We saw a demo where a developer from Florianópolis built a system that coordinates ten different "specialist agents" to do a full financial reconciliation. It uses the `context` package to handle timeouts across all those agents and `channels` to pipe the data between them. It’s so much more resilient and memory-efficient than trying to do the same thing in Node or Python. Host: Interesting! So Go is becoming the "manager" that tells the AI models what to do and when to do it? Guest: That’s a perfect way to put it. It’s the backbone. It’s the orchestration layer. Go manages the state, the self-healing loops, and the high-concurrency LLM streams. It’s just much more stable for production-grade AI than a scripting language. Host: I love that. Now, one thing that caught my eye in the conference reports was how much the Brazilian fintech scene is actually contributing back. You mentioned you’re at a major bank in São Paulo—how are you guys handling the move to 1.26? I imagine migrating a massive bank codebase is a nightmare. Guest: Oh, it usually is. Um, actually, that was the fourth big pillar of GopherCon LATAM: the new `go fix` tools. You know, Brazil has these very strict, fast-changing regulations from the Central Bank. We have to be able to pivot our code quickly. The 1.26 release updated `go fix` to use some really advanced AST—Abstract Syntax Tree—transformations. Host: Okay, explain that to me like I’m a junior dev. What does it actually do? Guest: (Laughs) Basically, it reads your old, "messy" code and understands the *intent*. Then it rewrites it into the new, faster patterns. For example, the new "range-over-func" patterns that make loops way faster. Instead of our team spending months manually updating legacy code, we ran the new `go fix` and it automated about 70% of the migration. It even inlined deprecated library calls. It’s not just making the code look prettier; it’s literally modernizing the architecture of a bank while we sleep. Host: That is a huge relief for any CTO, I’m sure. I mean, reducing technical debt automatically? That’s the dream. Guest: It really is. And I think that’s why the LATAM community is so excited. We aren't just using the language anymore; we’re stress-testing it at a scale that influences the global roadmap. The data our local companies provided helped tune the "Green Tea" GC for these high-throughput environments. It’s a very proud moment for the Brazilian dev community. Host: I can imagine! It sounds like the "Florianópolis vibe" was just as much about community pride as it was about the tech. Before I let you go, Mateo, what’s the one thing a Go developer should do tomorrow morning after listening to this? Guest: Download the 1.26 beta if you haven't yet. Try it on your heaviest service—just the compiler alone will give you a performance boost. And honestly? Look into the `context` and `channel` patterns for AI. Even if you aren't an "AI engineer," Go is going to be the platform where those agents live. Get comfortable with managing those concurrent streams now. Host: Great advice. Mateo, thank you so much for joining us and sharing all this. It sounds like Go 1.26 is going to be a game-changer. Guest: Thanks for having me, Alex! It was a blast. Host: Seriously, every time I think Go has hit its peak, the community just opens up a new level. The "Green Tea" GC and that 20% "free" performance boost? I’m definitely going to be playing with that this weekend. If you want to learn more about the sessions at GopherCon LATAM 2026, we’ll have links to the technical reports and the "Green Tea" benchmarks in the show notes at Allur.tech. Thanks for tuning into Allur. I’m Alex Chan, and we’ll catch you next week with more updates from the world of tech. Happy coding!