Back in 2005, my brother started his Bachelor's in IT when I was just a kid. During one of his assignments, he showed me how three different colour rings would show up when he pressed the run command, and it magically displayed three colour lines on a dark background. I remember this because when he showed me the code, it looked way too long for the mind of a kid.
The story is about how complicated and lengthy the coding process is, no matter what language you use. For a non-technical person like me, who has not studied programming languages, it looks too complicated to even try to edit the code. So if I have a website for personal use (like a portfolio) and I have to add or change something, it will either cost me to hire a web developer or require me to ask a friend to make the changes.
Now that same thing is being done using AI.
Recently, I came across a post on how Claude Code helps you manage work on your computer. Seeing that, I told Claude I was looking to build a portfolio website and did not know where to start. Claude ended up giving me a step-by-step guide on how I could do it. In the next prompt, I asked Claude how I could use Claude Code to have it do the work itself. In the next 30 minutes, Claude had a website up and running, while I only had to open webpages, sign up for accounts, and provide Claude with content. Within two days, the site was up on a live domain with working animations.
One of the web developers told me that aligning an image according to a requirement would take them two days of manual coding. Even given the skill set of the developer, the manual coding still took quite some time to get every image and piece of content aligned perfectly. Meanwhile, my portfolio website would open on mobile with the hero image aligned to the left, and it took me one prompt and ten minutes of Claude working (along with a git push) to make it centre-aligned.
The claims and the failures
I have been coming across stories of how AI is taking over jobs, and how many firms are firing developers and replacing them with AI agents. Even though it sounds like a good decision, in the long run it will not just cost firms expensive replacements in terms of money; they also risk data deletion and wrong decisions if AI is given access without supervision.
AI is an invention that will help corporations with better-performing employees, not replace them. We have already seen this happen.
In the case of PocketOS, an AI agent wiped their entire database and backup, exactly the kind of failure that comes from removing human supervision. And it is not only about loss of data, but also reputational loss. Recently, well-known newspapers like the Chicago Sun-Times and the Philadelphia Inquirer published a summer reading list for May 2025 featuring books that do not exist, simply because someone forgot to review the AI-generated list, and they ended up terminating their partnership with the editor responsible for the AI use.
What AI is actually good at
For the past few days, I have been using Claude to automate many of my regular tasks, build different websites, and get support for my PM role. The thing I have learned is that AI is excellent at operational tasks, like creating a static website with content provided, putting together business case templates, or automating workflows that do not require decision-making.
But when it comes down to whether a business case is realistic given the market situation, or what the firm's strategic alignment is, AI cannot creatively think around the problem the way a human with experience in the field can.
While working with Claude on website development, I had to keep going back and forth with instructions on what to change, where to place content, and what direction the website should take. Even though the AI kept advising on what could be done better along the way, in the end it had to be told what to keep and what to change. I am using a basic premium account, and after every few tasks it hits a usage limit that adds to the overall time it takes to complete the work.
The AI paradox
Now, keeping the above in view, the AI paradox becomes complicated. While it looks like AI can do every task much faster, and in some cases better than an employee, it comes at the cost of tokens, API spend, and creative decision-making. AI can create excellent presentations and business cases, or handle automated responses to customer feedback, but it cannot justify the case or presentation based on the company's strategy, market dynamics, or competition-focused feedback. This has always been done by the human. Human intuition cannot be replicated by AI.
The best real-life use case for AI is in the form of a copilot: the firm keeps the employee but supports them with the provisioning of the best AI provider, which is used to run normal daily operational tasks, keeping the API cost at the lowest while delivering the same amount of work in the shortest possible time.
The invention of Microsoft Office removed the need to keep hard ledgers spanning multiple pages or books, depending on the scale of the business. CAD replaced the manual creation of paper designs. In the same way, AI will be used more in a support role rather than as a replacement for human workers. The decision for employees and firms alike is whether they are going to add AI to their daily work or keep following the traditional systems. Whoever starts first will have a competitive advantage.
What this means for Product Management
In my own field of Product Management, that decision has already been made. The gap between a non-technical PM and a technical PM is closing too quickly to wait it out.
A year ago, a non-technical PM who needed a working prototype had to pull an engineer off their roadmap, brief them, and iterate through someone else's calendar. Today, the same PM can stand up a working website in two days with a laptop and an AI subscription, just like I did.
That changes what a PM is actually paid for. It is no longer about knowing how something gets built; anyone with a laptop and an AI subscription can do that. The value migrates to the parts AI cannot touch: knowing what to build, why it matters for the business, and what to throw away when the model confidently produces something that does not fit.