Why Coding Still Matters in the Age of AI Agents
Let’s talk about AI agents for coding, their impact on network automation, and how you should respond. TLDR: There are really good reasons to not give up developing scripting and coding skills.
AI Agents Write Code Today
People are definitely using AI agents to write code, and there is disruption being introduced by this. There are varying levels of performance with different LLMs and coding tasks, but most of the models are getting better, they are becoming more widely adopted, more models keep coming out, and they are becoming widely integrated into IDEs.
You're certainly hearing about “vibe coding” as well, and it has pros and cons; the code quality is less than perfect and the agents need supervision, but the ability to describe an application in your natural language to an AI agent is a game changer.
The bottom line is that agents are useful for writing code today, and the worst model you’ll use is the one you’re using today. This is a fact, a thing that is happening now, not just in the future. Using AI agents to write code is here to stay.
Don’t Give Up on Coding Skills
Now, if you’re working in network automation and not a fan of learning Python or other scripting languages, you may be tempted to throw in the towel on scripting. Aren’t these agents just going to take over coding and I won’t need to learn this?
The answer is No; there are multiple reasons that you should continue to develop your Python and other scripting skills:
(1) Adoption will take a long time: Avoid the trap of thinking “On day X, no one will ever need to write code again”. There is certainly disruption here, but AI presents yet another tool to help you write code, and learn how to code. AI models can be a very patient tutor in helping you learn coding and scripting.
(2) There is value for the long-term in understanding programming principles and details: One of the more common complaints about AI agents for coding is that they are not perfect, they can get hung up while developing code, and they can only handle a certain level of complexity. Being able to read the code will help you troubleshoot, and troubleshooting will help you develop more effective prompts to generate higher-quality code and scripts.
(3) Network Automation and Orchestration is more than just writing scripts: There needs to be understanding of process, workflow, and true system engineering across tech stacks for operations. Bringing that knowledge to your agents and understanding how they work (and the code they produce) will help you be a more effective “Automation Agent Boss”.
(4) Tasks and jobs are going to emerge that we can't yet foresee: One emerging role we’ve proposed via the Total Network Operations (TNOps) framework is the Operations Architect; these Automation Agent Boss tasks fit well in a role like the Ops Architect. We’re in the process of all this changing, probably very early in the process; we are certainly going to learn a lot along the way.
(5) There will always be a desire - and a market - to do things the Old Fashioned Way: This is more than sentiment; there will always be a need to troubleshoot problems with machinery of every kind, including AI Agents writing code.
On Timeframes
So how long will it take for agentic Coding to come into full bloom? What will this look like along the way? Here are some educated guesses on the timing of change.
(1) Short term: Agents still need supervision regarding their code output. There are a growing number of developers that use agents to write code who can provide examples of the imperfections: how the agents get stuck, produce odd outputs, and need supervision/help getting un-stuck. This will continue to get better over time. Just about anyone who has been experimenting with AI has noticed significant improvements with models over the last 12-18 months. Think about how the talk of hallucinations has reduced over the last 12 months; they still exist, but model developers are making progress here.
(2) Mid-term into long term: Adoption of new tools and real autonomy always take longer than you think. For example, we’ve had non-AI network automation tools for decades, but there are still barriers to adoption of network automation. It's still valuable to understand the underlying processes behind automation and coding so you can troubleshoot problems. I'm not a mechanic, but I know more than the basics of how my car works so I can identify issues. I'm not a carpenter, but I can still do some carpentry around my house. I can acquire skills that help me with a job without changing my basic identity.
(3) Long-term: I definitely expect disruption here. Code generation of any type (including scripting for automation) will be an area where AI will perform well.
Summary
Learning coding skills and how to use agentic AI to write code will be increasingly helpful tools in your tool belt across IT disciplines. Keep investing time in learning about them and exercising them.