Someone in your team hates AI?

I’m sure in your team or in your company, you have a skeptical Software Engineer. Someone that sais “I don’t use this stuff“, “I write my own code“, “AI is unreliable, I do it myself“.

That person might even be you.

Let me tell you a hard truth - I’ll not make it sweet - some engineers can resist the change, but the industry will not.

At some point, not knowing how to use this technology will be a competitive disadvantage.

Companies are starting to interview candidates and are starting to assess their AI usage skills. Employers are more interested in how you leverage the AI Assistant than in your ability to solve a LeetCode exercise.

In the past months, I’ve spent a lot of time with skeptics. This is what I’ve learned about them, and that’s how I helped them to change their minds.

First of all, the skeptics think “AI will write the code instead of me. I give it a prompt, it will produce a mergeable PR“. They try once, things break, the code sucks, and they reach the conclusion “uh… this magic thing is not that magic, so it sucks! I will continue to write code myself“.

That’s the main problem: expectations. We’re full of hype gurus on Twitter and LinkedIn saying that AI can automate Software Engineering. That’s not true - at least not yet.

The first thing to support a skeptic and start increasing the adoption of AI tools is a mindset shift. We need to see AI-generated code as a first draft, not the final code that can reach production!

It’s like asking ChatGPT to write us an email: we define our intent, let ChatGPT write the artifact for us, then we review it, fix it and send it over. It’s just… hundreds of times more complex!

Instead of writing an email, the agent needs to navigate through the codebase, make changes in multiple files, write tests, test their changes, look at the dependencies, search on the web, and so on.

Definitely not an easy task, so we need to help the AI.

How do we help? We bring the right context, give clear expectations, and expose the right tools. Our job changes from implementer to orchestrator.

Like a Tech Lead creating specs and assigning tasks to their engineers, while working with AI we do something very similar.

Thanks to better models, more mature tools, and methodologies like Spec-Driven Development, we can generate good enough code, which can then be reviewed, fixed if necessary and then merged!

Depending on your setup and your environment, you can expect between a 10% and a 10x performances improvmements (I know, that’s a huge range, but the context you’re working on makes all the difference in the world!)

A colleague of mine perfectly summarized his experience in this article, which I strongly recommend you read: https://ainativeteams.com/blog/i-was-highly-skeptical-but-now-im-a-believer/

In the end, if you have a skeptical colleague in your team, these are the steps I’d follow:
- He should change his expectations. AI is not a magic spell; it’s a (very powerful) tool
- He should treat AI-generated code as a first draft
- He should invest time and effort in learning context engineering and spec-driven development

Is there a learning curve? Sure! Will he struggle initially? Absolutely!

The engineers I’ve seen having great results using AI invested tens, not hundreds, of hours of training and practice, understanding how this new class of tools works before becoming proficient.

Buying a Copilot license is not nearly enough to get measurable long-term benefits!

If you have a skeptical person in your team, you should suggest that they take a look at this newsletter, or even look at my learning path, and invest one week to study and upskill. https://alfonsograziano.it/book#learning-path

I bet a beer on that: you’ll not come back once you master how to produce high-quality code with the support of Frontier AI Agents and Spec-Driven Development.

P.S. This newsletter has been entirely written without the support of AI (just Grammarly, as it spots obvious errors and missing commas XD). I deeply believe that AI can do a great job and can support a lot in writing, but this type of content is different. Is more like a flow of thoughts, something I’d share with you while taking a beer together.