Last month, I sent out a newsletter asking you: “What AI + PPC question is on your mind right now?”
Many of the responses I got were in line with:
“Do you still use scripts now that AI Agents have entered your agency?”
Yes. I still use (Google Ads) scripts. A lot.
My two main applications of scripts are:
First, Google Ads Scripts for monitoring and alerting.
Scripts are great for simple, repetitive tasks that do not require a lot of complex reasoning. Scripts are also great for automating rule-based logic based on numbers. They are also extremely reliable. They run, they do their thing, exactly the same thing and the same way every time.
Agents that use LLMs are much less reliable. LLMs are Large Language Models, not Large Number Models -> better with text than numbers. LLMs also tend to hallucinate and respond differently every time you ask them the same question. Therefore, agents make mistakes, just like us humans.
So, just like I use Google Ads scripts to monitor the changes made by humans, I use Google Ads scripts to monitor the changes made by agents. And I have them send me alerts when things go wrong.
Here are just a few of the many examples of the things I still like to automate via Google Ads Scripts (with no AI involved):
- Monitor account uptime, and send an alert if, for whatever reason, an account is down: https://nilsrooijmans.gumroad.com/l/account-down-alerts
- Monitor change history, and send an alert if someone, or something, outside of my team made a change in the account: https://nilsrooijmans.gumroad.com/l/change-history-alerts
- Conversion drop alert, sends an alert if a significant drop in tracked conversions occurs: https://nilsrooijmans.com/daily/google-ads-script-conversion-drop-alerts-and-a-big-thank-you
My second application of scripts is: software code as part of the agent skills.
These days, agents often have so-called “skills” to perform tasks. These skills consist of Markdown files (skill.MD) with natural-language instructions telling the agent how to approach a task. Example: “When adding new headlines, always spellcheck them before adding and make sure they respect the character limit.”
Skills can also have scripts. These are pieces of deterministic code that the agent can use to perform the task reliably. Example: “takes a list, returns per-headline pass/fail with reasons (over 30 chars, duplicate of headline #3, excessive punctuation, banned symbol, etc.). Pure deterministic check, no LLM needed.”
Without scripts, the agent skill would have to reinvent the deterministic parts on every run — e.g., re-deriving character-count checks, dedup logic, and API call structure from scratch in the model’s head. This is slower, more expensive, and far less reliable than simply executing code that already works.
So, yeah, scripts are here to stay.
Happy scripting!
– Nils