NAN124: AI and Trust in Modern Network Automation

The episode opens with host Eric Cho welcoming Sif Baksh, a Principal Solution Architect currently at Tines, a low-code automation platform. Sif shares his background spanning 15 years at DirecTV building data centers, followed by time at Infoblox working on DNS/DHCP/IPAM automation, and now at Tines evangelizing low-code automation for networking and security use cases. Both Eric and Sif reflect on meeting in person at AutoCon, which both describe as a rare and meaningful experience after years of digital-only connections.

The conversation traces the evolution of network automation languages, starting with Sif's early use of Perl and Tcl to automate ATM network trace routes at a carrier infrastructure company. Sif and Eric both note that Perl's flexibility — allowing multiple ways to accomplish the same task — made code difficult to maintain and share. Python's readability and enforced structure made it the natural successor, and Sif describes witnessing large enterprise teams adopt Python en masse about 10-12 years ago as a pivotal industry shift.

Sif then explains his initial skepticism toward low-code platforms, which he felt surrendered too much control, but how hands-on experience changed his view. He describes building in two hours a workflow that had taken a customer four months to build in Python with a custom JavaScript front end. He argues that the question is not whether you prefer writing code, but whether spending months building something yourself delivers more value than using a platform that accomplishes the same result in hours.

A significant portion of the discussion addresses AI's role in network automation. Sif describes using an AI story co-pilot within Tines to generate complete automation workflows from plain-English email descriptions. He also describes building an agent-to-agent firewall policy management system in 30 minutes that would previously have cost $120,000 per year in contractor fees or $500,000 in commercial software. He frames AI not as a job eliminator but as a force that elevates what engineers can accomplish, freeing them to tackle architectural improvements that were previously deprioritized due to operational burden.

Sif introduces his PENE (Prompt Engineering for Network Engineering) open-source framework, designed to teach network engineers how to write effective prompts rather than treating AI as a vague oracle. He argues that vague prompts like 'analyze the data' produce poor results, while detailed, context-rich prompts that mirror how a domain expert would think about a problem produce dramatically better outcomes. He observes that people with strong written English skills — such as lawyers or business executives — often interact with AI more effectively than technical engineers who default to terse, jargon-heavy inputs.

The episode also covers the broader career anxieties facing engineers in the AI era. Sif argues that the path forward is to use AI to accelerate learning in adjacent domains — he personally used Claude to develop a 90-day plan for improving his C-level executive communication — and to reframe AI-automated tasks as an opportunity to sell higher-value services rather than as a displacement threat. He gives the example of a firewall auditor whose annual reporting contract could be replaced by AI, but who could pivot to offering AI-assisted remediation, ticketing integration, and SOP generation as a more comprehensive and faster service.

Eric and Sif close with reflections on technology hype cycles, drawing parallels between current AI enthusiasm and previous surges around the internet, social media, and cloud computing. Both encourage listeners to start small with AI — solving one concrete problem first — and build incrementally rather than trying to master everything at once. Sif announces plans to expand his YouTube channel with content focused on AI-driven data center automation for newer engineers.