AI How will it take our jobs

This is a very fascinating topic and the consensus is that it is not a matter of if, but when. But no one seems to be talking about how this will unfold. I want to examine this a little in detail. Shall we start with small productivity improvements?

First, AI assistants will save time for professionals, leading to improved margins. Then with Improved margins, naturally it makes the entire field more lucrative. With less human input than ever, the same professional can output more work. Words will quickly spread about how much this job pays relative to the amount of work put in. A lucrative field will then lead to more entrants, which in turn creates more competition for the same job. As AI assistants do more tasks, the skill level required for the job decreases, and the competition degrades margins over time. Therefore to extract more profits, the professional needs a system that utilizes the AI to output more work. This overtime leads to the professional to be less hands on over time and more of an orchestrator of AI systems. So less of these jobs are needed because a single professional can handle more units of work than before. And sadly with this, these jobs will be gone.

This is how a profession can get wiped out. So where exactly are we on this progression and how fast does this progress? I have been told to never time the market and yet this feels like that is what I am going to do.

Looking at my industry, SaaS, we have created great tools for the past 20 years. But besides ChatGPT, I don’t quite see other AI assistants offering great value yet. Consider the typical software development process. A user need is usually discovered to be worthwhile to solve. From that, many discussions happen back and forth before a plan is put forth to break down the work into individual components. The product managers decide what to build and the engineers find a way to execute on that vision. The designers put together a mock UX/UI and the quality assurance engineers verify the product is free from bugs before releasing to customers. Once a piece of software is released, the customer support team helps users with any issues they encounter and report the feedback back to development teams for any bugs or improvements to make. This is a big team of people just to ship some software. How would AI come to the rescue here?

I think to crack this question, we need to start with the tools we use today. A big part of this development process is communication, think slack and zoom. Product and engineering managers spend most of their day in meetings and answering slack threads. Product decisions get formalized into business plans and technical specifications, which in turn get broken down into JIRA tickets and assigned to engineers. Designers use figma to mock up how the application will look and then Engineers will then write the code to push to a central repository, Github before getting deployed to some cloud system before reaching the users. Each role in the organization has its own tools to use and it is the person’s job to take the output of another person to create a body of work for the next person to use.

And it also seems like we are already making good tools for each professional role, which means we are still at a relatively primitive stage of white collar industries. I don’t see widely adopted toolchains that automatically handle the handoff of work between two professional roles.

So what is the vision I see?

To start, many conversations about what needs to be built happen in group discussions. This can be async such as slack messaging or meetings like zoom. From there, it seems natural to use AI to compile the documents necessary to enable the next stage of discussions. As the org incurs more meetings, AI should further refine the documentation and support planning to enable execution. From there, AI should generate the tickets necessary for engineers to take action on. Once the engineer has completed that unit of work, AI should update all the documentation because the code written may not match the original planning docs. Finally, when users are using that feature that was shipped. AI should be able to assist the users or customer support when they encounter issues. Does this feel like an AI project manager?

What are the tools AI needs to facilitate? Capture meaning from meetings, Slack, Zoom Business planning, technical specifications - Docs Breaking tasks down - JIRA Engineering support - Github Product documentation - Docs? Customer support - zendesk

So why is all of this done by a human today? Because when a human makes a decision on what work to do, he doesn’t specify the details on how the work should be done, and how much is considered adequate. The issue is how this work gets translated from inception to execution. Humans use language to think about how the work should be done and communicate that to other humans and machines. So LLMs will play a large role from language input into productivity outputs.