Google's new tool lets you build AI agents & workflows with a prompt

Wagwan hustlers,

It looked like it was going to be a quiet one this week… until Google launched their new workflow and agent builder yesterday (hence the late email).

This ones an action packed newsletter with plenty of gems for your work.

I also added some builders notes.

Google just launched Workspace Studio, a no-code workflow builder that connects your Google apps (Gmail, Calendar, Sheets, Drive, Forms) using AI prompts.

I haven't got access yet. It's rolling out gradually through January 5th. But I've been watching hours of live demos and following early testers on X to get a sense of what it actually is.

What is it:

Workspace Studio is essentially a Zapier-style automation tool locked to the Google ecosystem. You can glue Gmail to Sheets to Calendar to Chat without writing code. The AI component means you can type a prompt like "if an email contains a question, label it and ping me in Chat" and it builds the workflow for you.

It connects to some third-party apps like Asana, Jira, Mailchimp, and Salesforce, but the core value is tighter integration between Google's own products.

How it compares to n8n and Make:

Think of it like Canva vs Figma (design tools). Workspace Studio and Agent Kit are Canva: fast, friendly, and good enough for most people. n8n and Make are Figma: steeper learning curve, but the possibilities are endless once you know what you're doing.

The tradeoff is ecosystem lock-in and customisation. Workspace Studio ties you to Google apps. Agent Kit ties you to OpenAI's models. n8n and Make let you plug into hundreds of apps, swap models, build complex logic, and self-host everything.

Use Workspace Studio if: You live in Google's ecosystem and want simple automations between your apps without learning a new platform.

Use n8n or Make if: You need robust backend infrastructure, deep integrations, complex workflows, or full control over your data.

That said, if you haven't dipped into automation yet, this might be the easiest on-ramp we've seen. Workspace Studio lets you get that first win fast, inside tools you already use. For a lot of people, this could be their introduction to AI agents and workflow automation - and that's a win.

Bottom line:

Useful addition if you live in Google's world. Don't expect magic though. I'll do a proper hands-on breakdown once I get access.

Workspace Studio is rolling out through January 5th. Available on Business and Enterprise Workspace plans at no extra cost. If you don't see it yet, it's coming.

China's Deepseek released v3.2, and it's performing on par with GPT-5 and Gemini 3.0 Pro on most benchmarks… accept It costs 28 cents per million input tokens. GPT-5 costs $2. Claude Sonnet costs $3. That's roughly 7-10x cheaper for similar performance.

For most people, this won't change much. US models still win for everyday tasks. But if you're running agents, automations, or workflows that rack up API costs, the savings are massive. Deepseek is fully open-source, MIT licensed, and built specifically for tool use and agentic tasks.

Companies like Airbnb are already choosing Chinese models (Alibaba's Qwen) for their customer service agents because the cost savings are impossible to ignore.

OpenAI added a shopping research feature that asks follow-up questions, reads reviews, and builds buyer's guides so you don't have to. It's free for everyone and actually useful if you need to compare tools, gadgets, or gear without burning an afternoon on Amazon tabs. you can also use this feature for shopping software.

Ask a shopping question and ChatGPT will suggest shopping research automatically. Or you can also select ‘shopping research’ from the (+) menu.

I’ve been using Claude’s new Opus 4.5 model this week and its hands down the best AI model I’ve used for most tasks (just not for research).

It’s been working great for writing, and large context windows - I’ll run prompts with loads of documents and sources and it can handle them flawlessly. And it’s nuts at coding. I’ve just been using it to create fully functioning n8n automations.

I just describe what I want to automate, and Claude builds it for me. I’ve recorded a little demo of my process you can watch here (4min watch).

I started playing around with 'GPT-5.1 Thinking’ for copywriting after hearing good reviews - and it didn’t disappoint.

I’ve also been testing Gemini 3 Pro out a bit, and its great for deep research tasks, but overall I am not a fan at all.

You know when someone does something sleezy or lies to you once or twice and then from then on you just can’t trust them. This is how I feel with Gemini 3 Pro. It’s made some really odd mistakes while ive been using it on simple tasks on now I just can’t trust it.

Let me know if you guys liked the builders notes!! Could turn it into a seperate newsletter; one for news, one for practical implementation.

you can just reply to any of my emails with feedback and I’ll read it.

love,