Network Architect & Network Automation Owner #TechNerd #AI #python #Linux #LifeLongNerd
London, United Kindom
Joined December 2021
- Tweets837
- Following312
- Followers1.5K
- Likes1.7K
I’ve followed the Starting Strength methods for some time, training for strength rather than bodybuilding.
One overlooked benefit: mental toughness and durability - handling pressure, recovering from setbacks and keeping going when life gets difficult.
@SS_strength Thank you!
Taha Yusuf retweeted
Your CCIE Is Not Enough Anymore.
CCIE is still one of the most respected certifications in networking, and the knowledge behind it remains extremely valuable.
But AI is bringing new requirements to network infrastructure.
It's not just about connecting thousands of GPUs for large-scale training. AI inference, distributed workloads, storage networking, RDMA, RoCE, congestion control, low-latency fabrics, and AI data center interconnects are becoming increasingly important.
Traditional networking skills are still essential, but understanding how AI workloads behave and what they demand from the network is becoming just as important.
CCIE gives you a strong foundation. AI infrastructure is where you can build on it.
Have you started exploring these topics yet?
Everyone has heard of Kubernetes(K8s)
But what does it actually do in AI Infrastructure?
networkautomator.com/2026/10…
This article has received 11,000 views in the first 24hrs. The article is now also being referenced by Google 🎉🎉🎉
If you’re a network professional and you want to learn AI networking, please check it out!
networkautomator.com/2026/09…
networkautomator.com/2026/09…
I sometimes think of RoCEv2 as effectively band-aiding Ethernet for AI networking. There is nothing inherently wrong with Ethernet. It was simply not designed around AI networking model.
LangChain and LangGraph have become widely used frameworks for building agentic AI applications. LangChain provides the model, tool, and integration abstractions, while LangGraph provides the stateful orchestration layer that controls how an agent operates….
State tells the agent what it knows. Nodes define what work can happen. Edges connect that work. Routing determines which path is taken next. Loops allow the agent to act, observe the result, reason again, and continue working toward its goal.
My father has passed away…I will bury him today.
I don’t think I can ever express the feelings of losing a parent.
Just a test but wow….managing my ESXI cluster from my phone with AI agent feels amazing ……
just two years ago this would’ve been a sci-fi movie …
I always deliberately build ver 0.1 of my AI agents without MCP tool first. Replacing the Python dummy functions with MCP tools becomes almost trivial.
That separation gives me easier unit testing with fake tools..
LangGraph > GPU Tool > Mock Provider
#langraph #AI #AgenticAi
A British man has become the first person in the world to have a brain tumour removed via AI-assisted surgery.
This is why I got into AI/ML
I’ll like to remind everyone “inference” isn’t a single infrastructure profile, the traffic pattern and resource requirements can change significantly on the backend AI infrastructure depending on the serving backend parallelism architecture uses data, tensor, expert parallelism
66% of enterprises surveyed had shifted AI workloads from public cloud toward private or hybrid cloud, according to reporting on Cloudera’s 2026 Great AI Re-Architecture study of 1,500 enterprise architects and data leaders.
From an AI infrastructure perspective, an AI agent is fundamentally an inference workload, but with orchestration around it.
So an AI agent doesn’t introduce a completely new category of AI workload. It typically creates multiple inference calls, with tool/API/database.
#AI
Taha Yusuf retweeted
I’m thinking of starting **FREE live sessions on AI Infrastructure & Networking** soon.
If you are interested, **LIKE this post** 👍 so I can see how many people would like to join.
If there is a topic you want me to cover, additionaly **write it in the comments**.
Some of the topics we can cover:
AI Infrastructure • AI Training • AI Inference • GPU Architecture • HBM • PCIe • DPUs • SmartNICs • Scale-Up vs. Scale-Out • Data Parallelism • Tensor Parallelism • Pipeline Parallelism • Expert Parallelism • Mixture of Experts (MoE) • Collective Communications • AllReduce • AllGather • ReduceScatter • NCCL • CUDA • RDMA • RoCEv2 • InfiniBand • Ethernet for AI • Lossless Ethernet • PFC • ECN • DCBX • ETS • DCQCN • Congestion Control • Clos/Fat-Tree • Rail-Optimized Design • Dragonfly • ECMP • Flowlet Switching • Adaptive Routing • Incast • Elephant Flows • Tail Latency • NVLink • NVSwitch • GPUDirect RDMA • NVMe-oF • AI Storage • AI Storage Networking • AI Cooling • AI Data Center Power • Ultra Ethernet (UEC) • UET • UALink • SUE-T • MTU • GPU Utilization • AI Fabric Telemetry • AI Cluster Design
And probably many more as we go.
#AI #AIInfrastructure #AINetworking #ArtificialIntelligence #Networking
