John Crickett

YouTube video by Coding Chats What to do when AI gets expensive and other CTO challenges

Joined @johncrickett.bsky.social's Coding Chats podcast, if you like podcasts

Youtube: www.youtube.com/watch?v=z9Rr...
Spotify: lnkd.in/eNJw955r
Apple Podcasts: lnkd.in/epZnMjZz
Overcast: lnkd.in/eFNPGVvS

Tips for AI-assisted software development:

Boring tech gives AI superpowers.

AI coding agents performs best with tools, languages, and frameworks that have been around long enough to show up in its training data.

When use the bleeding edge, it hallucinates.

Tips for AI-Assisted software development:

Work in small batch sizes.

Humans and AI have limited short term memory. Ensure the task you're working on fits within your and AI's short-term memory.

When switching tasks start a new session, clear the context of you and the AI.

“AI is incapable of programming well thought out and complex code”

This is both true and irrelevant.

The goal is to write well thought out and simple code (which may solve a complex problem).

Is that working?

Did they take that approach to Scrum / Agile / Kubernetes / Programming Language / Interviewing Skills ?

I wonder if any other profession just has the "give it a go" approach to new tools :)

Did either company have a push for AI adoption?

If you use AI to help you build software professionally, has your organisation provided any training?

If so, what did it cover?

What was missing?

If not, why not?

That will depend on the scope of the user stories.

I tend towards stories that are as small as possible and will break those into subtasks.

I never have large commits or PRs. A lot of my PRs are single or double digit numbers of lines.

Build Your Own LLM Powered AI Chatbot | Coding Challenges This challenge is to build your own LLM powered AI chatbot. The chatbot will play a specific role and offer advice to the user. For example I’ve created one that is trained to behave like me and recom...

Software Engineers - need a project for the weekend?

How about building your own LLM powered chatbot?

codingchallenges.fyi/challenges/c...

Or one of the 80+ other real-world projects you can build to level up your coding skills:

codingchallenges.fyi/challenges/i...

Kiro: Agentic AI development from prototype to production Kiro helps you do your best work by bringing structure to AI coding with spec-driven development.

There's that and kiro.dev

I have tried neither. I would only expect a developer to work from a spec for a user story or part of one.

If you code with AI, and don't do Test-Driven Development, have you tried or are you going to try doing Spec-Driven Development?

What is the attraction?

What do you see as being a blocker to doing it?

Why do people feel the need to post stuff like this:

"So far only the uninformed and b-players are using LLM"

LLMs are way over-hyped, but attacking people is a weak argument.

Yes, I fear there's that real danger of making yourself ignorant.

I certainly agree building a project is the best way.

I'm still on the fence about using AI for it. I think it can work and I fear too many would lack the discipline to then read the code and try to understand it.

So it would end up just like watching a tutorial and copy and pasting the code.

If you use AI to develop software, do you vibe code or do you do AI assisted engineering?

Using AI To Solve A Coding Challenge Trying out augment code on the Spelling Correction Tool Coding Challenges.

In today's Coding Challenges I used Augment Code to develop a project in Gleam: codingchallenges.substack.com/p/using-ai-t...

If this doesn't look like an AI bubble, what does?

There aren’t enough good software engineering managers in most companies.

And this problem is getting worse.

The biggest, hardest and most common problems we have with software delivery are not technical, they're people problems.

AI won't fix that.

In software engineering, “it depends” often sounds like a smart answer.

But if it’s the only answer, it’s useless.

It suggests enough experienced to know that context matters, but not enough to be able to articulate why.

Next time you are tempted to say “it depends”, complete the sentence.

“If someone was a very good communicator but a terrible software engineer would you hire them?”

This is a false dilemma.

To be a good communicator, you need to understand the domain that you are communicating about.

"If you can't explain it simply, you don't understand it well enough"

Exactly, we now need to really focus on that loop - getting the specs, which has always been the hard bit that people try to avoid doing (because it's hard).

The hardest part of software engineering has always been writing good clear requirements in a natural language.

Using current AI requires we do more of the hard bit to automate the easier bit of writing code.

Finally, be prepared to replace members of the team who are not able or willing to learn and adapt to deliver quality software.

If your software engineering team is not delivering, a rewrite is not going the change that.

They can identify those areas that need most immediate attention and prioritise the changes, processes and procedures you need to turn around the development and keep your business moving!

That will only be fixed by developing the team's skillset. Find someone who can lead, mentor or coach your team and have them work with the team to incrementally improve your software.

It wasn’t the language, library, platform or architecture that was wrong – after all they managed to deliver something and you have customers getting value from the software – it was the people, processes or procedures that resulted in poor quality, late software.

It’s not true. If you throw away your product and start again, you’re setting your business back months, maybe years, all whilst losing revenue and market leadership.

And there’s no reason to believe that your team is going to do a better job this time around.

I’ve met many tech executives frustrated that their software teams were failing to deliver.

In each case they’ve been advised by their software engineers that the answer was to rewrite all their software with some awesome new approach, technology, language, library, platform or architecture.

🧵👇

Is anyone using Zig in production?

If so, what and why did you pick it?