Alfred Simon

Working on AI tools for PPC people - building PPC OS and other tools.

[Alfred Simon @AlfredSimon]

Important context this happens when you want to buy this from your workspace account. I don't want to buy this via my personal account.

[Alfred Simon @AlfredSimon]

Everything that is wrong with Google in a 3 min video
I'm sure this can be done a bit easier or maybe I'm just stupid

[Alfred Simon @AlfredSimon]

@lorikmor but 100% of people who don't use toothpaste also die, so toothpaste is not the cause

[Alfred Simon @AlfredSimon]

@lorikmor You need to post about why toothpase is not good for your IQ and some other health take. You get 1M in a blink of an eye

[Alfred Simon @AlfredSimon]

@franzbruckhoff By the amount of CVEs from october I think not just vibe coded apps had bugs or exploits

[Alfred Simon @AlfredSimon]

@kultekin probably was vibe coded

[Alfred Simon @AlfredSimon]

What does it even mean to have a good Google Ads AI Agent? It comes down to 4 simple principles.

In order to be able to define the above we need to define what is even a good Google Ads account and good account management.

A good account with proper management should tick the 4 points below:

1. It supports in achieving the well defined business goals
2. It keeps doing that on a longer time scale
3. It spots chances of growth and risks on time, and it is easy to adapt to them.
4. It builds on what already works.

Now all you need to do is make sure that the AI that you use knows about all 4 and he can check up on all 4.

Let's take an example for each

1. It supports it

The agent should know what the overall business goal is and what role the Google Ads account plays in the whole plan to achive that. Ideally it should be able to see the actual data behind this.

This is your context layer everything to know about the business, goals, challenges etc.

2. It keeps working

This point is mostly the management. In this case you need to make sure it has access to all secondary data points that help him make decision to achieve point 1.

This is also the layer where you need to make sure it has enough tools to be able to do that and here comes the CRITICAL part of control.

What it can or can't do without you looking at it.

3. It spots and adapts fast

This layer is mostly outside data that the agent can't see from the account or from the business context layer.

Competitor activity, pricing changes, performance of other channels, communication between you and the client etc.
Probably one of the hardest layers to get right.

Example: You can't blame claude when it doesn't push harder, when your biggest competitor is out of stock and nothing in the account tells him that. He needs to know about that.

4. It builds on what already works

This is your SOPs turned into skills/instructions layer. Tell it how to do certain things based on your decades of experience which LLM's don't have, because they were trained on trash.

It should know how to make an LP that works based on your tests and way of working. How you spot trends in the accounts etc.

These are the core vision points behind PPC OS also.

[Alfred Simon @AlfredSimon]

@dedene well this is also a nice presentation of that it's our responsibility to help the model :D. I saw the movie and love it, but in this context

why put two red buttons that do the same thing next to each other? We need to make sure they can't make a crucial mistake this simply.

[Alfred Simon @AlfredSimon]

@omarchy when m4? :D

[Alfred Simon @AlfredSimon]

@yongfook I really recommend Orca. The best combination of Terminal and IDE

[Alfred Simon @AlfredSimon]

This is what people with 0 Google Ads knowledge sound like when they talk about ads and marketing

[Zack Korman @ZackKorman]

This is what non-cybersecurity people sound like when they post about cybersecurity.

[Alfred Simon @AlfredSimon]

@ZackKorman I will steal this and frame it in 10 different ways thank you so much :D

[Alfred Simon @AlfredSimon]

Google Reps in the AI era are just nothing more than you emailing with a Gemini model that has little to 0 understanding about your account or business goals.

It made 0 sense before AI and it makes now even less sense.

[Alfred Simon @AlfredSimon]

@hammer_mt @rickmanelius thanks for the reply, makes sense!

[Alfred Simon @AlfredSimon]

@hammer_mt @rickmanelius It might be a stupid question, but aren't they running them locally? Or you think it's a built in feature for chinese models to feed the goold old communist party?

Asking as someone with 0 experience using GLM

[Alfred Simon @AlfredSimon]

Model evals and performance benchmarks for Google Ads. Something I'm working on as part of a big PPC OS update

In the current version of PPC OS I'm still missing a proper eval system, which is now what I'm working on.

There are a few things I want to be able to monitor with this:

1. If I change a skill's instructions did it get better or not?
2. Which model is better at Google Ads related problems? Is that Opus 5.5 or Deepseek V4? At the moment there are no official benchmarks for this only for coding stuff
3. Which CLI/harness is better for Google Ads work.

First one is done for the account auditor skill. A fake client account with 24 planted problems, plus a clean account to catch false alarms.
50 runs so far on Opus 5.5, Fable 5.1 and Codex, and all of them find all 24, so I need harder cases or all of the models are just as good :D

A system like this gives us a chance to alert our users if they really need to use a different model, or tell them if its worth buying Codex or not.

It might also be that one model is good for feed related issues and one for search terms. Super awesome to see how it works already, but it's still early days, but this will add a lot of value to PPC OS both for us, who make it and for our users.

[Alfred Simon @AlfredSimon]

@mattshumer_ What are apps that you build like that? just asking to learn, since for example an app I'm working on I catch claude doing stupid stuff regularly so doing what you say would be bad for me

[Alfred Simon @AlfredSimon]

@BerntMuurling Yeah realised that now :D crazy stuff

[Alfred Simon @AlfredSimon]

I was today years old when I found out people are trading on the potential launch price of Anthropic stock.

A speculation on a speculation :D that is based on a vibe I guess.

And there are actual platforms for this

[Alfred Simon @AlfredSimon]

If I ask your during onboarding "Do people at your company use AI" and you reply "yes, our whole company is using Copilot" I've got some news for you :D