Raw LLM Responses

I disagree. He bases his argument on an obviously false equivalency, obvious at least to those who want to see reality, and who don't choose to deny an unpleasant truth. He does this by comparing AI to what he claims is a human doing a multitude of tasks; requirements, design, customer interactions, integration of software, etc. Unless you're a company run by you and a couple of your friends, one person doesn't do this, and you shouldn't expect one AI to do it either. It involves allocating different tasks to different people or teams, and likewise allocating tasks to AIs trained to do different things. For example, when I did years of software design and development as a contractor across different organizations, private and government, there was a team that met with customers to develop requirements, then either they or another team would meet with hardware and software teams to develop specifications. Then another team, usually the more experience developers would create a general integrated design. Then the software tasks were divided into different parts, and the individual programmers assigned these tasks would first submit their designs, and when peer approved, develop their code. This software was then integrated into a general system, with a quality team testing at each phase, first test approving the individual code, then the integration, and then after implementation (live). Often the testing team wasn't involved until integration, with the developer doing the unit (his program),testing. When applying AI, an analogy is creating roads with no heavy equipment. Crews of people dig with shovels, then they use flat shovels to flatten the surface with buckets of water and bags of gravel or sand to cover the road, then they carry buckets of hot asphalt and empty them. Then use flat shovels to beat down the asphalt. Once construction equipment is developed, over 80% of these people are not needed. With software development, substitute those using the equipment: bull dozers, dump trucks, road rollers, etc with a greatly reduced team that will assign AI tasks and integrate them. The requirements gathering in complex environments is better suited for people to perform, but these aren't developers anyway. The specifications can be accomplished by a specifically trained AI that will calculate the hardware and software to accomplish the task and the division of tasks and how it will be resource distributed. A smaller team (two or three people), like the drivers of equipment in road construction, will review these specifications. Then another specifically trained AI will input the human edited specifications and create the design of the software in both flow charts and English statements ( puesdo code). This small team will review and edit if necessary, then each of the parts in the design will be submitted to AI for development. These components will then be integrated and tested by a small team. A specific AI can be trained to receive the requirements, and then create test scenarios to test the integrated software to assure it meets these requirements, and even conduct the tests itself. A small team will review the results. When there are bugs or integration issues, in some cases, a small software team will solve them without AI input, and in other cases, they might use a specifically trained AI to troubleshoot (offer suggestions) of potential reasons for the bugs. On an average project I worked on, and there were many over the years, there were over fifty people across the different tasks in small endeavors, usually it was over a hundred people, and sometimes it was in the hundreds. With AI, this staff can easily be reduced by 80% as in the case of the road crew analogy.