Hello and happy Tuesday, Thursday! 😰

As we near the end of 2025 and overwhelm sets in with all our unmet goals; we've decided to make this issue something of a personal retrospective on the past year.

Make a brew and let’s jump in… 🫖

🔎 Preview: In today's issue

• 😶‍🌫️ Kieran's view on vibe coding is changing
• 🤔 Matt's thoughts on AI agents
• 📺 James's hot takes from Asia

It's been quite a year for changes in the AI landscape and at times I've felt paralysed by all the shiny objects and feeling of being left behind.

Here are some things I've realised...

I’m a laggard and that's okay

I’m slow to try new tools. Always have been.

Keeping up with every new model, framework, or AI product feels impossible, so I’ve mostly stopped trying. I don’t consume much “latest AI news” content anymore. Instead, I wait for the dust to settle.

My rule of thumb is simple:

If people I trust are still using the same tool or model six months later, then I’ll pay attention. Most things don’t survive that long. The ones that do are usually worth learning.

This approach has saved my sanity and, ironically, led me to better tools.

Recommendation:

Pick two or three people whose judgement you trust and copy their tool choices with a six month delay. You will miss very little and waste far less time.

My view on vibe coding is evolving

For a long time, I was sceptical of vibe coding.

Early attempts felt fragile and frustrating. Things broke easily. Debugging felt opaque. I didn’t trust what was happening under the hood.

That’s started to change.

Over the past couple of weeks, I’ve been experimenting with Claude Code, and it’s shifted my perspective more than I expected. It just works. And it opens up a new world of building possibilities.

More importantly, it unlocked a new realisation: I don’t just enjoy building apps. I enjoy product managing an AI to build apps for me.

That’s a different skill entirely, and one I find surprisingly satisfying.

Recommendation:

If you have written off vibe coding, revisit it with a single focused project. Treat the AI like a junior developer that needs clear instructions, not like magic.

AI is best when it's teaching you

I’ve settled into a workflow where I pair coding agents with a separate AI chatbot.

Claude builds and Gemini teaches me (I use Gemini because I like the way it writes, but use whatever model you prefer).

I use it like a personal tutor and ask relentlessly basic questions about what I’m seeing. I look at the output and try to understand why it exists and what problem it solves.

I don’t want to write the code myself. It’s still extremely complex. But I do want to understand the moving parts of the systems I’m creating.

Recommendation:

Use one AI to do the work and a different one to explain it to you. Ask naive questions on purpose. Ask it to give you analogies related to something you understand well.

The opportunities are real

It’s an incredibly exciting time to be AI and software literate.

I constantly hear real stories, not hype, of people in non-technical roles spotting a problem, building software to solve it, and eventually leaving their jobs to run that business.

This isn’t easy. Most attempts still fail. But the gap between idea and working software has never been smaller, and it’s shrinking fast.

Importantly, the more removed your job is from the world of AI and new technology, the more opportunities there are. Software companies are drowning in tools already. Trades, local government, healthcare, legal services, hospitality... this is where software can really make a difference.

Check out my 5-step day job escape plan blog post.

Recommendation:

If you have a problem you complain about regularly at work, try building a tiny internal tool to solve just that one thing. Real leverage starts small.

My favourite AI agent

...is beautifully simple and real-life.

It helps me to stay on top of all the dates I need to remember for my kids' school.

It's a Zapier agent that:

1. Triggers when I label an email in my Gmail account with a calendar agent label
2. Reads the email and extracts events I need on my calendar
3. Checks to see if they're already on my calendar
4. Adds them if they're not there and sends me a confirmation message

It honestly saves me a lot of annoying admin!

Recommendation:

Find small but annoying admin things you do in your personal life. Build a Zapier agent (it's very easy, just like writing an email) to fix it.

Create With Conference 2026 tickets are on sale now.

Last year's conference sold out so why not grab your early bird ticket today?

I've been head down teaching Agents and Automation over the past months. Here's some questions I've been posed that I thought were great!

How to decide what to build?

When you start, you want to build the clever thing. But the agents people actually use daily? They're boring.

My advice

1) start with your calendar and try to find a repetitive thing you do...then think what do i need to prep for this.
2) Go to the templates page of the tool you are using.
3) For more fun, browse Apify and think of interesting data sources you could use.

​
Either way, think about your personal productivity first.

For example, my most-used agent is a 'chief of staff'. Nothing fancy. It reviews my day each morning and surfaces what matters from Calendar, Gmail Notion and Slack. It saves me 15 minutes of context-switching and triggers me with some actions to take.

That's it. But I use it every single day, which is more than I can say for the elaborate systems I've built and abandoned.

The pattern: build for frequency, not complexity.

Architecting agents

There's no exact science here, but the principle that works is the same one that works in all software development: break things into logical components.

For agents, think in roles. A researcher. A drafter. An editor. Each with a clear job.

Yes, the models are getting powerful enough to handle enormous prompts and reason through them. But control matters. When something breaks (and it will), you want to know which component failed. When you want to improve output quality, you want to tweak one role, not untangle a monolithic prompt.

Separation of concerns isn't just good engineering. It's good debugging.

Keeping up with tools

Someone asked how to stay on top of all the new tools launching. Honest answer: even I struggle, and tracking this space is literally my job.

Yesterday I went back to Apple Notes for my task list. I'd built an elaborate system in Notion with an AI project manager handling tasks, updates, the lot. It worked. But the cognitive overhead of maintaining it outweighed the benefit.

Sometimes the right tool is the boring one you already know.

The meta-lesson: don't confuse experimenting with tools for actual productivity. They're different activities. Schedule time for exploration, but don't let it colonise your actual work.

📺 James in Bangkok

​

What happens when you combine walking robots, limitless demand for intelligence, and a region hungry for technological change? You get a front-row seat to the future of AI in Asia.

Live from AIMX Singapore, host Seng Woei Yuan sat down with James. Fresh from hosting their first Asia event in Bangkok, James shares his observations on what makes this region different. In this episode, they discuss:

• Why the "AI bubble" narrative doesn't apply to Asia
• The surprising presence of embodied AI and robotics at AIMX Singapore
• Asia's energy advantage and openness to change
• A prediction: AI supercomputers in space, powered by solar
• Why traditional apps and operating systems may be on their way out
• The massive opportunity for AI adoption across Southeast Asian enterprises

Whether you're building, investing, or just curious about where AI is heading – this conversation offers a refreshingly optimistic take from someone who's seen both sides of the world.

▶️ Watch Video​

💡 Want to share your product with 6000 readers?

​Click here to help us to keep the lights on and become one of our sponsors. We have a consistent CTR of over 40% and an audience of tinkerers, builders and pros.

🧠 Tiny AI Tip

A friendly, clear and firm tones yield better, more direct results

❌ Fix the grammar in this. NOW.

Aggressive tones can lead to overly cautious, pre-canned, or less helpful responses as the model tries to de-escalate. Politeness can sometimes result in chatty, less direct answers.

✅ Please review the following text for grammatical errors and suggest corrections. My goal is to make it sound more professional and confident

This is direct, respectful, and provides context ("goal is to make it sound more professional"). (Greg Isenberg)

🔊 On your radar

Keep your finger on the pulse of what we think is hot in the space right now 🔥

⭐ Visual editor for Cursor Browser: design meets code, in one window​

👉 Why it matters: You can drag and drop elements on your live site, tweak props and styles, and describe changes by pointing and typing. The agent then updates the actual code for you. It's getting close to a Figma (even Bubble.io) approach.

⭐ ChatGPT 5.2 quietly raises the bar on real work tasks​

👉 Why it matters: The latest model handles long documents far better, spots issues in screenshots, and can generate real Excel files with formulas. For you, that means faster audits of PDFs, cleaner dashboard reviews, and working spreadsheets without manual setup.

⭐ n8n announce Chat Hub Beta: a single chat for your n8n automations​

👉 Why it matters: If you use n8n, Chat Hub gives your team a simple chat window to run workflow agents without exposing the workflows themselves. You can set “chat only” users, plug in multiple AI models, and keep costs and access under one roof. It’s in beta, so expect changes, but looks very interesting.

PS...

If you want to learn more about AI and love a deadline we've partnered with 100 School's Agent Bootcamp. All Create With members get 20% off all bootcamps.

💭 Final Thoughts

We were wondering how we might get to work in the future? This was one idea

There will be no newsletter next week as it's Christmas but we'll be back the week after.
​
With good vibes and festive cheer ✨
​
The Create With Team