Software Development Changed

After taking some courses on Claude’s academy and reading/watching some blogs along the way about agent development, these is a FAQ of questions and its answers I had at the beginning that I am writing here in case they can help others in my situation.

Agent development is a must-to-know?

Yes, it is getting better every quarter and if you are not using it for coding then you will miss out lot of good and bad things, but mostly good. If you are in IT you must know already that there is one constant in this career and it is that we have to learn to learn new things. I see this as an another wave of transformational changes, like Web and mobile development.

I don’t know if it is a bubble yet, but even if it is I think is worth the time to learn how to use it for making you more productive now.

Has really software development changed forever?

Yes, taking the last point of the previous question. This seems to me a technology that it is here to stay, even if the bubble burst, the servers and current tech will stay and it is useful as it is for coding. On the other hand, this is the worst version of AI for coding that we have so it will get better in the most optimistic version.

I feel coding will not longer be manual, not considering exceptions of some industries, legal restrictions, companies not trusting AI, etc. Most jobs are okay using AI for software development and honestly some companies can see a return in reduced time-to-market, for example.

I don’t think it is all benefits though, like junior roles being impacted. However, we will talk more about this in later questions.

Are the courses on Claude academy worth the time?

Mostly yes, they are short and easy to follow. Just remember that Claude is not the only way of coding with agents since Open AI Codex is also very popular in the software community. I started with Claude since it was the de-facto AI for coding some months ago, but like any AI topic, it now has changed to be less or same as Codex.

Here are some recommended courses on Claude Academy though:

What are the most important concepts to know?

What are good skills and agents to have?

Depends. Please don’t copy without understanding the skill purpose to your project. There are million of skills if you google them. However, this is a repo I like from Addy Osmani: https://github.com/addyosmani/agent-skills

Which model should I use?

Laboratories want you to spend money on their latest models for almost everything, because it is the most expensive and intelligent model they have. However, one does not really need the most powerful model for every single task.

My advice is to pick the model based on the level of complexity of the tasks. Example: finding vulnerabilities in code with Fable 5.1, but for coding Opus and Sonnet are fine for now. Remember, these models change almost per month so it is true at this moment.

However, this does not mean that you won’t challenge yourself finding ways to use the most intelligent model, since here is where the real advance is. My suggestion is to try to think outside your confort zone of tasks and use the latest models for some hard things you can think of.

What about code quality?

I am glad you are like me and care about code quality. There are some people that argue that the quality decreases just because there is lot of new code that is humanly impossible to review and keep up. Others say, you need to keep checking code but based on a clear definition of what can pass without a human review and what really needs you or the team to take a look and review the code.

I like the second approach, there are trivial changes that are easy to omit review. Medium changes that a good glance and understanding of the change is enough, but there are complex changes in terms of length and complexity that really need our attention before to even think of pushing these changes to production one day. Be mindful of the type of change and talk to the team about your standards with AI.

Will AI make us lose our jobs?

Ok, now we are entering the grey area and a bit controversial topics of coding development. I don’t think AI is and will make us lose our jobs for now. I like this quote from Peter McCrory:

“So far, AI is both skill-biased and labor-augmenting. It complements domain expertise. It relies on humans in the loop to direct and evaluate the most complex work. And it rewards AI proficiency. Model capabilities are improving fast, but remain stubbornly jagged.”

However, the speed of models improving is very hard to keep up and amazing at the same time. If there is one thing that AI is relatively good at is at coding. Claude for instance, let you plan, create specs, design, code, test, deploy and monitor with AI. At a first glance, it may appear AI doesn’t need us but the opposite is true since AI for now cannot proactively do tasks by itself since it requires a person or team to form a good output of any phase.

On the other hand, there is an impact on junior roles that I can see happening or decrease in hiring pure software roles. Companies are realising that productive is growing as long their teams keep up with the knowledge to use these technologies.

Is AI really so good/bad?

Honestly, I don’t know. Some people argue that AI is basically a bubble created by companies that produce it so they can obtain revenue (Internet of Bugs). Others argue, that it is almost an existential threat to humanity (Rutger Bregman).

My view, is a little bit of both. Yes, companies and specially Open AI only cares about revenue, but it is also true that can be disastrous if it is used bad and without governance.

I think it has definitely some benefits for society like drug discovery, but if used wrong it can be a mass cyber security weapon. Like any tech, it is not good or bad by itself, but the purpose that we give it is what it makes it beneficial or dangerous for society.