Another Developer's Thoughts on AI

After a couple of months of heavily using both Co-Pilot and Claude for development use I have a few thoughts. I know…I know another person jumping on the hype train talking about AI, but hear me out. There’s such a hype train around AI and how it’s going to take our jobs I think it’s worth putting out a more responsible view point to cut through this crap. It’s a great tool and I have no doubt that it is here to stay, but I’m not jumping on the bandwagon that this is the best thing since sliced bread. For a few reasons

1. It’s the most capable junior developer. It can output more code than any one person - but it can make so many bad decisions
2. Cost - I’m not sold that the current pricing structure is going to hold. Enterprise users are costing a lot more than the consumers and people are maxing out their usage all the time
3. Will we get a positive ROI from a business standpoint?

“A good software developer will be able to make more good software… a bad developer will produce more problems. These tools are nothing more than a major reflection of the user.”

Capabilities #

Anyone who’s used AI knows how capable it can be. That being said, I would warn non-technical users that there’s a lot more to writing software then just what you see on the screen. There’s a roadster welding together in the backyard and a Toyota Corolla. There’s a learning curve to using AI but I believe that these models will amplify a person’s capabilities. A good software developer will be able to make more good software… a bad developer will produce more problems. And this is the major warning that I have seen many developers are trying to warn people.

Personally and professionally I have been able to ship new features and personal projects much faster. It has helped me with skill gaps - I hate writing css and crafting a good design. I know what looks “good” but it takes me longer to get there. With AI I have been able to skip a lot time required to get something decent but this has brought up other issues in the process. These gains are not universal and there was a lot of iterations before the end results the public or users saw. My fear is that those who are good with AI and understand it will lead others to think that this is a net positive for everyone, even those who do not understand the basic of software engineering.

For those who have spent time with these models will be able to point out the gaps. Like outputting code that “works” but will create a mess without guard rails. Or gas lighting you into thinking your solution is perfect when it’s not. Too just the times where the responses are coming back strange because Anthropic is throttling your output due to high overall usage. I’m sure talking with anyone who’s used these models can tell you which ones are better suited for different tasks. Not all models are the same.

Cost #

I have become that software boomer who argues with people online about a few things…and cost is the major one. At this point we’re still in the subsidizes costs of this race. None of these AI companies are making a profit and they’re keeping their consumer costs down to attract users. Once they have the base and people can’t or won’t code without this tool they’ll start to raise their prices. This is the old playbook in business and I’m surprised that top executives are not talking about this in public. For companies that are not tech, but need tech to operate they shouldn’t care about what OpenAI or Anthropic are doing on a technical level - only that their product can help them produce more results in whatever form gives them a return. But it seems like everyone on Wall Street just wants to spit out the same saying.

But at what point does the cost of these tools outweigh the return they’ll give us as developers and in turn business? Right now I think there will be a balance of token/prices per employee that is getting worked out in the market and I don’t believe any CEO who spouts these numbers that would gain for higher usage *cough cough Nvidia. I want to hear this from companies that are in other industries who are using the tools and have to really balance the cost per return, and I’m sure we’ll see more of these in the next few years.

But what happens when that gets organically figured out and then these AI companies start increasing their prices?

ROI #

This brings the two points together. How can we as developers work with the business to use these tools to get a return on their “investment”? Excluding the external factors like price increases I think a lot of teams will have to adapt. In the heads of an experienced developer these tools can give us the ability to do more, but we can’t restricted by red tape. This is a personal problem of mine but there’s nothing worse then being part of a team and being so narrowed down that you have to kick parts over the fence to finish. You give me this great tool that allows me to get more done but then sidelined. I get it - we have a team for a reason to people who specialize in certain aspects of the stack can ensure that is done correctly, but this can be so overdone. And it can be one of the biggest roadblocks to leveraging AI.

I think this is where the concept of smaller teams keeps getting brought up. AI will not make you a better design, but it can give a back end developer a good starting point and vise versa. It won’t completely take one specific role but allow a smaller team with different specialize talents like design, fullstack developer, and even data person the ability to output much more than a larger team that is broken into “traditional” roles. Again, this is just a hunch and we’ll see how this shakes out.

Overall, I think these models will be a net positive for development but there is a lot of snake oil salesman out there right now. These tools are nothing more than a major reflection of the users - good developers can output great tools, but a lot of crap can be produced as well.