How we use (and avoid) AI in production

If you’ve worked in or around software lately, you’ve seen the wave:

🧠 Claude Code, CoPilot writing code

💬 Fireflies, Fathom taking meeting notes

⌨️ Cursor helping shape entire dev flows

And honestly? We use them too. Daily.

But here’s the truth nobody wants to say:

This week, we’re pulling back the curtain on how we actually use AI in production projects, and where we still draw the line.

Here’s what we’ll cover:

⚙️ When AI gets it right

🚀 Where we still say no

🔑 How we strike the balance

At the weekend, I was standing on the roof of Coughton Court, a National Trust property close to home.

600 years of history, stone walls, and stories of civil war and the Gunpowder Plot.

Quite the location for Halloween weekend, and as tomorrow is the 5th of November, it couldn’t have felt further from modern-day and the world of AI and automation.

But weirdly, it made me think about how every generation faces its own version of “the future.”

The tools change.

The fears repeat.

And that’s what this week’s issue is about; not the hype, but the real use of AI in our work today.

When AI gets it right

There are three areas where AI tools have genuinely changed the game for us:

1. Initial code scaffolding

   Claude is great at generating structure, saving us time on boilerplate or quick setup tasks.

2. Speeding up documentation

   Fireflies, Notion AI and similar tools help summarise calls and meetings. Perfect for capturing key points or writing initial drafts.

3. Debugging, in context

   Tools like Cursor can flag possible fixes or suggest approaches based on the context we’re already working in, great for reducing mental overhead.

Nugget #1: AI isn’t replacing good devs, it’s giving them fewer tabs and better starting points.

Where we still say no

But… AI isn’t magic. And it’s not always safe.

We don’t trust it to:

❌ Ship production code without a real review

❌ Interpret complex business rules

❌ Make architectural decisions

❌ Replace testing or QA

When AI guesses wrong (and it does), you don’t just get bad code. You get confidently wrong code, and that’s dangerous.

We’ve seen it suggest logic that looked right, but missed edge cases that cost real hours to fix.

Nugget #2: AI can do 80%, but if the last 20% matters, you need humans who know what’s at stake.

How we strike the balance

We treat AI like an intern with superpowers:

• Fast

• Always available

• Occasionally brilliant

• Often needs supervision

That means building workflows where AI assists, but never owns the outcome.

We’ve also started baking some of these tools into our internal platforms, not to automate everything, but to speed up the parts that shouldn’t take brainpower.

It’s not about hype. It’s about helping people focus on what actually matters.

Nugget #3: The smartest teams use AI to reduce friction, not responsibility.

What can you do today?

🔹 Audit where your dev team is losing time; is AI helping, or hiding shortcuts?

🔹 Pick one tool and trial it in non-critical workflows

🔹 Ask: What’s the riskiest thing we’re letting AI touch right now?

Use AI where it earns trust, not just where it’s trending.

The future’s exciting, but only if it actually works.

P.S. If you’re trying to figure out where AI fits in your stack, we’ve probably already tested it (and broken it). Let’s save you the trial and error.

Need a fresh perspective on your software? I’m here to help. Book a 1-1 call with me, and we’ll figure it out together.

Do you know somebody else who could benefit from The Automation Advantage?

Before You Go…

How did you enjoy this email? I really value your honest feedback.

🔥 It was AWESOME!

🔥 Very good!

🔥 Okay!

🥶 Needs work!