In my experience it's really good at making you feel that you understand things.
Then when you actually dig into the code, there are many things that are not like you'd expect.
When you've experienced that a few times, you stop trusting that the agent gives you the full picture - for good reason.
When I review AI generated code I generally find so many flaws that it makes it hard for me to believe that those who are not reviewing their output are not just fooling themselves. Maybe not all the time, but quite often.
One such recent example was an SSO simulator for a local env. Instead of using a cookie to remember who was logged in, the agent remembered the last log in a variable, assuming the the next requests would come from that login.
This snowballed into our tests, where later agents had created helper tools for working around the SSO simulators statefulness.
Things improve drastically however if you spin up a second session and ask it to adversarially review everything that the first session produces (this goes for everything: not just code, but also design, planning, and explanations).
This works even better if you use models from different families to do so.
I have, on a lark, reached 20 to 30 adversarial sessions a few times. For some tasks, the sessions will just never converge on anything that yet another session won't find fault with, recommending an alternative already ruled out by another session. Even if all the reasoning in between was documented, the new sessions will endlessly claim to find flaws in past reasoning.
The fact that i would use 99% less dependencies, and overall have a WAY more simple deployment story. Also a typed language is always better than the PHP way, where you have functions that are hard to reason about, and edge cases llms miss easily.
Exactly. I feel that agentic programming is largely turning what would have in the past been actual understanding into some kind of fluffy superficial pseudo "understanding" that's only barely good enough when assisted by AI.
My boss tried to make me interested in agentic programming by showing me how "productive" the creator of OpenClaw is.
All I see is a fast food factory. Sure it's interesting that you can produce so much so quickly. But what is produced is endlessly uninteresting to me. It doesn't help that all the harness building is just SEO in disguise. Trial and error.
I don't doubt that agentic programming makes it possible to produce faster, but it also comes at a great cost for me, which is that I largely don't enjoy my job anymore. If I was given the choice between that and halving my salary I would have chosen the latter.
> I doubt that they will go back to where Sony are now.
I agree. However, I do think they would get some positive attention (and some accompanying sales) if they were to backtrack and announce a console more like the 360.
It feels doable if they care to do it. Physical media should still be viable for holding all the game data for a while longer. Blurays can manage up to 128 GB, and I think the average game install size is ~60GB right now, giving most games some room to grow.
The biggest issue with a strategy like that is that they're, like you said, pushing digital-only hard already, and they're also trying to save money, so the idea of spending more money to make future consoles with disk drives, and to make disks, is unlikely to appeal to them.
It is a shame, though, because it seems like the Xbox 360 will have been widely viewed as peak Xbox until the end of Xbox.
I don't disagree that this device is very likely too expensive to sell well.
But! The price is not insulting. You can built a slightly faster PC for a little less, but that PC would be ~10 times larger, it would be louder, it would lack features like HDMI-CEC and good wifi/bluetooth. It really wouldn't compare for living room usage.
In order to get anywhere near the size of the Steam Machine, you'd have to exceed its costs.
Isn't CEC available for all in-tree GPU drivers using DP-to-HDMI passthrough? (although I do imagine Nvidia would still be preferred for a custom build, perhaps with a USB CEC injector)
Then when you actually dig into the code, there are many things that are not like you'd expect.
When you've experienced that a few times, you stop trusting that the agent gives you the full picture - for good reason.
When I review AI generated code I generally find so many flaws that it makes it hard for me to believe that those who are not reviewing their output are not just fooling themselves. Maybe not all the time, but quite often.
One such recent example was an SSO simulator for a local env. Instead of using a cookie to remember who was logged in, the agent remembered the last log in a variable, assuming the the next requests would come from that login.
This snowballed into our tests, where later agents had created helper tools for working around the SSO simulators statefulness.