• Human review seems to be the bottleneck for AI-generated code, it’s time to rethink the traditional PR workflow. No or less human in the loop.
• Kubernetes is a good platform for long-running workloads like microservices and batch processing. But not ideal for short-lived ones like agents as computers today, which require strong isolation, fast snapshots/restore, and rapid scaling up/down. Time to build a new platform for agents with micro-VMs. But GPU support is still a challenge.
• Personal agents are the next product phenomenon. But it should be deployed locally.
• Local AI will be the key enabler for personal agents, ensuring privacy, low latency, and offline capabilities. Also the big thing for AI infrastructure going forward.
• Every company should hire someone working on AI for workflow automation just like the way they hire for DevOps or SecOps.
• With AI, everyone is a researcher today, since coding is no longer the main bottleneck.
• AI4Everything (AI4E) is the moat for almost every technical company - the differentiator is how deeply they embrace it.

Stadio Diego Armando Maradona.

At hiverge.ai, we’re building Hive, a distributed platform where large numbers of self-improving evolutionary agents for algorithm discovery and optimization. We mainly focus on three domains:

It’s 6 a.m. in Cambridge UK, and I’ve been lying awake since 3 a.m., the reason is just because Meta released a new product called Muse, it’s a long-running personal AI assistant inside a confidential VM with long-term memory and the ability to improve the foundational model over time. This is quite similar to the AI system I want and am building now, which makes me excited but also really anxious.

Feel the energy of life, no AI.

[Slides]

The CNCF Technical Advisory Groups (TAGs) play a vital role in shaping the future of cloud native. Explore the TAG Workloads Foundation with us! This session will present the mission, scope, early initiatives, and subprojects of TAG Workloads Foundation, focused on defining and advancing practices and standards for cloud native workload execution environments and lifecycle management. Attendees will learn how this TAG supports the CNCF’s technical vision, why workload execution is critical to adopters, and how community members can get involved to help solve real-world challenges across container platforms, schedulers, orchestration systems, and more. Join us to help shape the next phase of cloud native maturity—from fundamental runtime environments to future-forward workload patterns.

Hiking Catbells in Lake District, UK. Raining the whole day …

Dynamo is a high-throughput, low-latency inference framework in a distributed setup. Recently, I’m working on optimizing the Dynamo system, copiloted by our own product, the Hive, an evoluationary AI agent. Router is one of the key components in dynamo, especially like the KV-aware router, which seems like a good starting point to work on after talking with the Dynamo team.

Review the Goals of 2025

Looking back 2025, I set three goals (all about the work) at the beginning of the year, it turns out that the results are not that satisfying:

It was my birthday last week, too busy with work to write anything. But I think it’s a good time to imagine what I want to achieve in the next few years, or say, my long-term vision for 2025. Especially when AI is changing the world so fast, it’s important to have a clear vision of where I want to go.