AIn't All That Chief

Same homework everywhere

People are missing one huge thing about Al security and that is similar output. These Als are fine tuned so that they ensure the answers it provides are coherent and something that the user expects to receive. This is to say that the answers it gives can be different but they always tend to the same answers because they are all trained on very similar data sets and they all follow similar principles. This is a big security threat. Why? Because as more and more companies mandate that developers use Al (and to be clear, they should) the developers will all use the same code. This means that there are the same vulnerabilities across all codebases. Think of the log4j exploit from 2021 turned up to 11, but this time, it isn’t just one part of the code that is vulnerable to the same exploit, it’s the entire codebase. Hackers will learn the paterns of output of the Big 4 AI models - Claude, ChatGPT, Deepseek & Gemini, and once they do, they’ll pretty much have your source code. Then, mix this with ever more accelerating development will only lead to an increase in cyber threat.

Another major security threat is social engineering. Google released their Veo 3 tool a number of days ago and the level of realism is absolutely ludicrous. In recent years, we have been made aware of how to spot social scams but I’m sure there have been a few times where you thought

“Wow, this is pretty convincing.”

but once you see Veo 3 you will realise that it is only going to become harder and harder to spot reality from a scam. If you have not already seen its capabilities, then I implore you to check it out.

Both of these issues open the doors to more and more problems that need to be solved by people and those people will benefit greatly. I can see a rise in solutions like Java’s SonarQube (which is a code ‘smelling’ tool used to enforce good coding practices) only the new ones will report on “too much Al code” or “Al similarity implementation” issues. Tools used to determine if something is A.l. generated or not will be very sought after.

The Scalability Issue

The next thought I’d like to address is likely a concern for everyone - Al job replacement. There are many reasons I would like to give for this not happening, but, with the limited allotment I have, let me propose a scenario and then reason about why that won’t happen.

I’d like to raise that, if a business currently produces 10 things with 10 employees and it costs the company £100; but, they can produce 10 things with 5 employees + Al and it will cost them £70, then they will make that change. Now each employee’s productivity is higher, but, the total output remains the same. Is the business just going to sit on this? No; business strives for growth. So the business then rehires the 5 staff. It now produces 20 things with 10 employees + Al and it costs £140 - a £60 saving.

The flaw to this thought, regarding supply, is that you could argue:

“Why doesn’t the business then just scale the AI agents instead of humans where it is not a 1-1 relationship between the human and AI, but rather a 1-many relationship?”

Then you run into a scaling issue. If we extrapolate to the extremities to pain the picture, we can see if there is a 1 person to 1000 agents relationship, then when something goes wrong, identifying the issue and fixing it would be very difficult. Not only that, but you then have 1000 agents interacting with each other, being fed bad input but executing on it 100x faster than a human, meaning, the speed at which the issue spreads is colossal. For the Managers reading this, your job is difficult enough with 1s or even 10s of reports - image 1000s without critical thinking.

Furthermore, to the scalability point, I have two anecdotes that help paint my point. Firstly, a very senior member of a project that I am on, recently showcased their enthusiasm for an Al tool called Replit, and its ability to enable them to produce a sophisticated-enough dashboard, written in Typescript, demonstrating the speed at which Al enables us to work. They exclaimed that

“What would normally take a human, 5-6 weeks to create, Replit created in 5 or 6 hours.”

Impressive I thought, but then in the same breath he said

”… it wrote over 26,000 lines of code…”

to which I was skeptical. Coming from an engineering background, I know that the most valuable solutions are often the simplest and most basic - not the ones with the most lines of code. Simpler solutions are typically more extensible and maintainable. At a glance, 26,000 lines of code for what was only a dashboard Ul seemed very verbose.

And secondly, I have noticed that a non-trivial number of junior developers coming into the industry have relied so much on Al during their university courses that they posses no capability of writing code themselves. Even now in industry, when questioned on their code during a code review, they simply reply,

“I don’t know, ChatGPT wrote it”.

And, I would postulate that more and more mid-level to senior developers will start to lose their skill as they rely on AIs more - whether due to tighter deadlines or other pressures. Given this, I suspect that in the future, the gap between high performers and low performers won’t necessarily be due to high performers becoming better and better, but rather because low performers will continue getting worse and worse.

Conclussions & Actions

Through all this, the message is clear. You will not be replaced but how you operate will. When an automised and mechanised world was born from the industrial revolution, many jobs were displaced. When the information and digital age came along, many many more were affected, and each time, these people just had to adapt and learn how to operate in their new environments. And, the fact is, even after these events, fortunately or unfortunately, you and and I are still taking the tube each day to our 9-5s, so this will very likely be the same and we too must learn to adapt. Being aware of and using these new tools everyday is a total must. Be intentional and critical with all usage. Only by this, will you will begin to see how your role in the future might be.