I predict we are less than 12 months away from the pendulum swinging back to hand coding but dragging ai pair programming into the workflow. This is why I don’t worry too much about this.
I can predict this because I’m building such a thing for myself as are many others. It seems to be where the tao of slop is going.
At work I've spent the last few weeks building a big feature that involves streaming the output of a 3D engine to a web client and wiring in HID from the client back to the engine. Basically, it's a simplified version of Google Stadia.
Writing code has never been the bottleneck. System design, testing, integration, CI/CD, etc. These have always been the things that were the most time consuming and thought intensive. Those haven't changed as much.
However, if you don't think there are productivity gains by several orders of magnitude when actually writing the code, you're either using it wrong or you were the fastest dev in all the land.
I'm not trying to be rude. Is this new to you? I've had all of this plugged in myself for all of 2026 (at least). The LLMs use of APIs shouldn't be surprising once you see decent ones in action in general. I'm not sure how this would impact my thoughts on whether the pendulum will swing one way or not.
This is new to me (last 6 months or so). What I’m observing is people growing lazy. I think it will take a while for it to swing back because momentum is still in the wrong direction.
Yeah, reality is this fall/winter is when we start seeing the shift from "engineer guides claude on their laptop" to "PMs write tickets and everything else is hands off until pressing the merge button".
I actually agree with what you said! the subscriptions we are paying today is only a tiny fraction of what it actually cost AI companies like Claude and Codex. either it will shift into running those agents in your own setup or pay an absurred number. anyways as you said the pendulum will swing back for sure!
The ship has sailed. For personal use, maybe, but supply and demand are going to dictate that there is going to be a massive ramp in memory and compute production, and when the prices inevitably normalize due to the increase of availability, there will be a push to use local inference. It goes back and forth, and unless there is a strong roadblock put in place, the current generation of models are already powerful enough that incremental improvements in the run up to increased hardware availability will mean that these tools are here to stay. Even if draconian laws were put into place, the prevalence and availability of open models and the tools to train them means you can't put the genie back in the bottle.
Definitely not. Look at what Deepseek costs. AI companies spend a fraction on inference, it is not expensive for them to run these models like it is to train them. Coding by hand is not coming back.
Large enterprises are paying unsubsidized metered rates. I work at a firm with around 700 employees. The p90 engineer is spending $2k/wk.
Our business-side employees have only begun using agents, and their spending accelerating. They're claiming their ROI is massively positive and have an enormous willing to spend.
Prices will probably go down because of the emerging price war between Anthropic and OpenAI, but there are many firms willing to spend.
I can predict this because I’m building such a thing for myself as are many others. It seems to be where the tao of slop is going.