I’ve been actively working with AI tools (ChatGPT, Github Copilot, etc) for about two years now – both in my day job as an Cloud/DevOps engineer and as an off-hours amateur app developer. Here are my thoughts thus far in short form:
1. AI tools are extraordinarily helpful for specific tasks in which you know exactly what the outcome needs to be. I do a lot of Terraform, Github Actions pipelines, and Python scripting to automate deployment and configuration. If I can conceive of a configuration step to be handled by py script, and I can accurately construct a prompt to get it from an LLM – then I can get something useful that 100% meets my needs within a few prompts. What used to require probably an hour or more of online research digging through documentation, Stack Exchange, Reddit, etc is now quickly handled by a well-worded prompt. It cannot be understated how much time and effort this conserves, allowing more attention to be spared on other projects. As long as I know exactly what I need, I can get it quickly.
2. AI tools are great for analyzing errors and suggesting solutions. I spent a lot of time recently configuring Azure WAF rules and exclusions. This required significant WAF log analysis to zero in on problem traffic and understand why exactly certain parts of legitimate traffic were being blocked erroneously by WAF rules. In the past, this type work work – much like item 1 above – required a lot of research and digging through documentation, not to mention frequently needing a support ticket for more clear advice on our specific situations. Enter AI tools – and now I can get extremely fast error analysis along with suggested solutions within minutes. Logs showing a WAF block? I can provide log details to the LLM and get a quick (and usually accurate) analysis of what exactly caused traffic to be blocked, and how it should be remedied – whether requiring an exclusion at WAF, or requiring a code solution to mitigate a previously undetected vulnerability or poor code practice.
3. Vibe coding is – at least right now – a great recipe for disaster. Vibe coding is fun – I’ve tried it with humorous and interesting results. But woe betide the engineer who actually tries vibe coding for enterprise use, or anything requiring security / stability – the results from my experiments with vibe coding for enterprise-grade solutions (which I have NEVER actually used by the way, just experiments for fun) have created the most wacky and unsecure stuff I’ve ever seen – not to mention convoluted in most cases. I have no doubt that “vibe coding” will improve significantly in the coming years and may even someday be sufficiently stable to be used safely for certain business purposes – but anyone trusting vibe code right now is playing with fire.
Categories: Products
Austin Thomé - Cloud & DevOps Engineer 