• Human Infrastructure
  • Posts
  • Human Infrastructure 392: Writing Code with LLMs, Automation Vs. Software Development, and More

Human Infrastructure 392: Writing Code with LLMs, Automation Vs. Software Development, and More

AuthorAuthor

Drew Conry-Murray & Ethan Banks
March 13, 2025

THIS WEEK’S MUST-READ BLOGS 🤓

I like how this post addresses the good and bad parts of using LLMs as a coding assistant. They’re not magic, and you’ll have to put in some effort to get useful results. Simon shares his experiences by offering a simple list of lessons learned, which include “You have to test what it writes!” and “Tell them exactly what to do.” He then provides more details about each lesson. He notes “Using LLMs to write code is difficult and unintuitive. It takes significant effort to figure out the sharp and soft edges of using them in this way… .”  While Simon is addressing software developers, his lessons should apply to network engineers dabbling with Python or network automation. - Drew  

Claudia de Luna puts a stake in the ground with this statement: “Network automation is software development.” It’s a statement that some, perhaps many, network engineers don’t want to hear. It doesn’t mean you can’t embrace network automation if you don’t embrace software development yourself. But if you want network automation and aren’t ready to DIY the software side, you may need to hire a developer. Read Claudia’s post to get her full argument, but I’ll leave you with this from the blog: “Not every network engineer needs to become a developer or a scripter but I believe every network engineer is going to need to be familiar with much more of the software development and server tooling to be successful in the future.” - Drew  

Hans writes about his experiences handcrafting packets with Scapy as a vehicle for experimenting with half-open TCP connections. What’s a half-open connection? One that hasn’t been fully opened, as implied by the name. For a TCP connection to be fully opened the 3-way handshake must be completed. The sender initiates the connection with a TCP packet that has the SYN flag set. The receiver responds with the SYN & ACK flags set. The sender confirms with an ACK.

A common attack is to tie up system resources via a “SYN flood” where the supposed sender (or senders if a DDoS) bombs the target with SYNs in an attempt to saturate the target’s connection table with half-opens. Why half-opens? Because the attacker never completes the open process with the final ACK of the 3-way handshake. In a past life, I admin’ed a device that sat in front of our Internet-facing, high-value reverse proxies in our DMZ. One of the device’s party tricks was to proxy TCP handshakes and discard SYN floods, protecting our mission-critical reverse proxies.

Time has marched on, and modern operating systems have gotten better at protecting themselves from SYN flood attacks. How good? Hans shares his findings. - Ethan

Collaboration is generally regarded as a good thing within an organization. But why? And how can you make it happen? Regarding why, this post argues that collaboration supports innovation. If your org sees benefits in innovation, collaboration should be a goal. As for how, the post provides some prerequisites for fostering collaboration, and then describes four pillars required to build a collaborative culture. It also shares concrete ways to inject opportunities for collaboration in an organization. If you’re looking to foster collaboration within a team, across departments, or even org-wide, this post provides a useful roadmap. - Drew 

This blog walks through the steps to configure dual-stack PPPoE on a Fortigate FG-60F. It includes step-by-step instructions, plus GUI screenshots or CLI where applicable. - Drew

MORE BLOGS

  1. Networking Fundamentals (epic list from Ivan Pepelnjak of his “fundamentals” content - bookmark it!) - ipSpace

  2. BGP: Setting Up and Peering in the DN42 BGP Network Using a MikroTik Router - Melson Mascarenhas

  3. Kubernetes@Home – what do you do if your ISP changes your IP addresses? - Vegards Blog

  4. The DevTools Ceiling: great as Open Source, AND a Terrible Business - DX Tips

  5. Becoming a good engineer - Madmeg

  6. Vibe Coding (experiments in AI prompt engineering) - Thinking About Software, Life, The Universe And Everything

  7. The End of CLI as We Know It (more AI prompt engineering) - Generative Programmer


Every second counts. Web performance isn’t just a technical task—it’s a business imperative.

In this Catchpoint webinar, you’ll hear from experts on advanced web optimization methods, tools, and strategies to help you enhance performance, deliver exceptional user experiences, and implement continuous optimization to stay ahead in 2025.
Key takeaways:

  • Discover how to optimize your website to achieve peak performance.

  • Understand how to leverage advanced tools and techniques such as real user monitoring (RUM) and WebPageTest (WPT).

  • Learn how to analyze web performance tests and uncover hidden performance issues before they impact your users.

  • Explore real-world use cases that demonstrate the business impact of optimization.

TECH NEWS 📣

Per El Reg, Broadcom “has added around $1 billion to quarterly VMware revenue in a little over a year. How did it do that? As The Register has often explained, Broadcom stopped selling standalone VMware products and now only sells bundles of code and support under subscriptions that are more costly than Virtzilla’s previous licenses.”

Specifically, VMware customers are being upsold the VMware Cloud Foundation (VCF) bundle. About 70% of VMware’s top 10,000 customers have bought VCF. - Ethan

The key technology at work here is photonics. Company Xanadu has built a photonic quantum computer called Aurora “that can operate at scale using several modules interconnected through fiber optic cables.” Aurora can scale and handles both fault tolerance and error correction effectively.

Most quantum computers use superconducting qubits paired with microwaves to process data, but have to be cooled to near absolute zero to function—that has made quantum out of reach for all but dedicated facilities willing to spend the money. Aurora is a first baby step towards broadly accessible quantum computing. “By using light-based, or photonic, qubits instead of microwave or superconducting qubits, [the Aurora] team created a light-based system that uses networked photonic chips. This makes Aurora inherently connectible, as fiber optics make up the basis of the global networking system.”

I’m pretty sure that once we pair AI with quantum, the singularity occurs. - Ethan

This is an in-depth look at the industrialization of phone scams, where callers (typically operating out of big call centers in countries such as India) trick people into sending them money. It’s a Canadian publication so it focuses on the costs to Canadians ($530 million in 2022), but phone scammers are a global scourge. - Drew 

You know the drill. - Drew  

MORE NEWS

FOR THE LULZ 🤣

RESEARCH & RESOURCES 📒

This 21-item digital book bundle is a cornucopia spilling over of Python titles, including Mastering Python Networking by Eric Chou, host of the Network Automation Nerds podcast. The Humble Bundle model is “pay what you want”. Pay less than $18, get fewer items. Pay $25 (the suggested amount) or more, and the extra supports a worthy cause. Looks like this bundle is available until the end of March 2025.

Top tip. You can feed PDFs (such as your shiny new Python collection) into Google’s NotebookLM and ask it questions. You’ll get back answers with citations so you know where in your book collection to dig in and find out more. - Ethan

Sidekick (local-first LLM) - johnbean393 via GitHub
https://github.com/johnbean393/Sidekick

From the README. “Chat with a local LLM that can respond with information from your files, folders and websites on your Mac without installing any other software. All conversations happen offline, and your data stays secure. Sidekick is a local first application –– with a built in inference engine for local models, while accommodating OpenAI compatible APIs for additional model options.”

There’s a good bit more in the README that explains techniques & other capabilities. Well worth reading if a project like this intrigues you, especially if you don’t love using the AI models and engines supplied by the Broligarchy.

Sidekick runs on Macs with Apple silicon. I am very interested to try Sidekick. - Ethan

This is a bit more than a list of research papers on eBPF. It’s a curated list of papers that you can filter based on a variety of eBPF-related criteria. - Ethan

MORE RESOURCES

  1. Age and cognitive skills: Use it or lose it (neuroscience research) - ScienceAdvances

  2. NLnet Foundation (funding for the open Internet)

  3. MeetingBar (free, Mac, join meetings in one click)


Calling all Wireshark users! 

Register now for the SharkFest’25 US conference - June 14-19 in Richmond, VA - to learn from the best in network analysis. The conference, focused on sharing knowledge, experience and best practices among the Wireshark® developer and user communities, will feature a keynote by Vint Cerf, recognized as one of "the fathers of the Internet". Meet the Wireshark core developers, network with your peers, level up your skills, build your professional network and much more! Don't miss out on what past attendees have called "the best conference in the industry" and "a privilege to catch up with the Wireshark community."

Registration link and more details at: https://sharkfest.wireshark.org/sfus/registration-options/

INDUSTRY BLOGS & VENDOR ANNOUNCEMENTS 💬 

The king is dead. Long live the king! We’re all rooting for you, Lip-Bu. Maybe the board will give you more time than they gave Pat. - Ethan

TL;DR. The OnCall & Incident products have been merged into Cloud IRM. Changes to show up in the Grafana Cloud UI soon, and it shouldn’t break anything if you’re an OnCall/Incident user. - Ethan

Cisco’s Vijoy Pandey says, “Today, we're announcing the AGNTCY - an open source collective building the critical infrastructure for AI agents to work together. We invite the entire AI and infrastructure community to join us in this town square for collaboration. Your expertise and contributions are essential for creating an open, interoperable foundation that serves everyone. Cisco, LangChain, and Galileo are the initial core maintainers, with Glean and LlamaIndex joining as contributors.”

This is interesting positioning for Cisco, but makes sense considering this context Vijoy goes on to provide. “While single agents can handle specific tasks, the real power comes when specialized agents collaborate to solve complex problems—just like humans do in the real world. But there's a fundamental gap: we have no standardized infrastructure for these agents to discover, communicate with, and work alongside each other.

An open, interoperable Internet of Agents is the best path forward to accelerate innovation and create the most value for all participants–from builders to operators, developers to consumers. Just as the original internet connected computers and the web connected information, the Internet of Agents will connect AI systems across vendor and organizational boundaries, and technical frameworks. Without this foundation, we're essentially trying to build the web without RPCs, HTTP, DNS, or TCP/IP.”

I haven’t thought too hard about this yet. But my gut reaction is that this feels like a big—if low-risk—bet for Cisco. They stand to benefit, as do the other participants, if AGNTCY becomes a de-facto standard for agentic AI communications. If it doesn’t work out? They can absorb it. - Ethan

MORE INDUSTRY NOISES

DYSTOPIA IRL 🐙

TOO MANY LINKS WOULD NEVER BE ENOUGH 🐳

LAST LAUGH 😆