Not long time ago, the corporation environment complains about how many 'No, sir' they were receiving from their dev teams, because of the high demanding, high asking, no priority pool, no conversation, inflate backlogs and everything that you can think about the daily basis of a regular company. Then the AI came up to solve this, to bring productive, high-speed delivery, reduce the technical bottleneck, and all excuses that non-technical fellows would love the hear.
No, sir
This moment where we still adapting about the many frameworks were propagating, migrations from local to cloud, scalable designs, discussion as "microservice vs monolithic", or any other tech topics that you can think of. Our demand to be on the latest topic was huge, and manage the tradeoff between new and old framework, introduce new bugs and maintain legacy code could delay the implementation of new features on the systems, pick a new task from the backlog just to cleaning it.
Because of these back-and-forth conversations among all the actors that compose a regular (let's call old just to laugh a bit) old team structure: devs, devops, PM, scrum master, QA team, manager, and so on, we constant said each other 'No, sir... we cannot implement this in the moment. Please, add this task to our backlog'.
Then the team's structure and ways to address and news idea swift to a fast pace, no boundaries, limit only by the numbers of token allowed... if you have this limit!
Yes, sir
What no one could think of (I mean, few folks over the world advise this, but not important right?!), is that tradeoff is the lack of quality vs more PRs, the bottleneck now is the high volume of non-checkable code, hidden bugs, extra "features that goes along with your prompt, just because the model found a similar vector to your question, even that the context / business implication is complete different. And, as any coder know, is harder to validate and reviews PRs with thousands of file changes.
Now, the fast-paced mind has a tool that only respond 'Yes, sir... as you command', even that don't do it as you asked, but it doesn't limit you as well. They can implement, test the idea as they have always wished, only that they don't know what is happening in the background.
Also, productive is measure by completeness tasks, but without detailed and clear criteria, the bar is too slow, but, hey, you are moving faster than we have ever had! Critics choices are not in the table anymore, there are no conflict ideas, because there is only user prompt idea. You have a bias servant applying the commands you give.
Do you allow to proceed with the next phase, sir?
Why, sir
Recently in a conversation among some students, I couldn't emphasis enough about how important we are as a human layer in the new time of tech, how our understanding about the world and humanity are a gate keeper of the responses for the AI. I mean, they are not evil or good, they are tools that can mimic humans' thoughts, behaviors and decisions. It's like have a small piece of yourself in a very specific point of the time (fun when you know that matrix weights in training are called checkpoints).
When we bring back the human conversation and interaction, after seeming our services stall, forgotten because they didn't solve any real problem than our ego to delivery something, even the lower minimal thing. Question mark like:
- Would we have experience to question?
- What is actually a good question?
- Who should be questioning it?
Not sure if hope is the world, but I am looking forward to the imperative speak tune reduce, and the question raise again.
Top comments (0)