This sounds suspiciously like a capacity story masquerading as a safety story.

I feel like there is some use case planned here that isn't to be known about until it's way too late to do something about it. Or this is a very serious bubble. One of the two or some really horrible blend.

> The Postgres community's position on query hints has been, shall we say, firm. The official wiki page on the subject states it plainly: "We are not interested in implementing hints in the exact ways they are commonly implemented on other databases. Proposals based on "because they've got them" will not be welcomed."

Are we reading the same article?

If you are going to be handing tokens to AI agents on weird openclaw contraptions, you should try to use the fine grained variants. My GitHub account spans 3 organizations with wildly differing policies. The fact that classic tokens are even still allowed blows my mind a bit. You should be required to manually opt in each organization at a minimum.

Principles driven development (we will never or always do X regardless of context) typically comes off as a petty ego trip. The point of the technology is to serve some kind of downstream business. Most people who download Postgres are seeking to solve a real world problem, not to demonstrate their ideological purity.

If a gigawatt class DC suddenly needs to take its sensitive IT loads off grid, it could be designed with load banks on site to stand in for the IT load. These would be exceptional use only, so the specific cooling technology (obviously we want to boil water) is not much of an ecological concern. A gigawatt will vaporize ~100 gallons of water per second. How long until the grid can adapt? Five minutes? That's not exactly a heroic amount of water for these projects.

As a retail electricity customer, I vastly prefer whatever is going on in their operational context.

I live 10 miles from the "border" between ERCOT and MISO and I've got a lot of experience with both, particularly in disaster scenarios. ERCOT is a total shitshow when it comes to the edges. The fuel mix in MISO is pretty terrible for the environment, but it is also unbelievably reliable. The Mississippi River is not to be underestimated in its logistical impacts. The price of natural gas drives the daily fuel mix proportions, but it doesn't affect the available base capacity and its financial status (fully amortized for over a generation). If natural gas gets expensive in ercot and the wind isn't blowing, there isn't anything else to give. MISO has 2-3x the coal and nuclear capacity that ERCOT does.

The "intelligence" is clearly there now. Trying to measure it seems pointless. I can't shop for hammers at the hardware store and sort by the quality of finished products they would produce. That is clearly an insane ask, but that's approximately what is being pushed for with these models now.

Domain specificity (harness & environment) is where the magic happens next. I intentionally use a slightly less powerful model to help reveal weakness in how I've exposed the domain to the model. Having capability reserves available dramatically increases confidence around a project like this. If the customer starts to complain about some edges, I can crank them up to gpt5.5 for target scenarios. If I'm already on 5.5 there's nowhere else to go. I'm up against the wall.

Ownership and responsibility are the new currency for the engineering staff. Willingness to implement these tools and then own the consequences of their use is what leadership is looking for. They want their cake while they eat cake, and they will keep those around who enable something approaching that experience. Owning the side effects of LLM use is more challenging than our own natural output because of the radical volume increase and unfamiliarity with low level details. However, I argue it is still possible. It has always been significantly more expedient to poke holes in someone (something) else's work than it is to perform that same work. And, the executives know this. They leverage this capability too.

The relationship between the business and the development team has been tenuous at best. I've rarely seen a technology team that was properly subservient to the business that ultimately signed their paychecks. I every case I have personally experienced, it is was like a hostage situation where the business owners are in constant terror of the technology people screwing them over in some infinitely nuanced way they or their lawyers could never understand. Many business owners are looking at this technology as a way out of the hostage situation. They noticed a window that was left unlocked. They are going for it right now. Whether or not they will succeed in their escape is a separate matter. Whether or not them being held hostage was justified is also a separate matter. It really helps to keep these things in their own lanes.

What kind of domains did you have in mind?

> This is not a political post

Why is this rage bait on the front page of HN?

I still don't understand how caching helps me very much. I must be misunderstanding it because I thought the user's prompt (which is the biggest variable) necessarily sits prior to all of these token intensive tool calls. How can we cache the reading of codebase if the prefix is always moving?

If you can make all the STUN servers fail from the perspective of the clients, you could hypothetically force them to use TURN servers that are more centralized and easier to spy on. STUN negotiates pipes n:n. TURN is closer to n:1.

I'm seeing a ratio of around 10:1 in my usage. A vast majority of the tokens consumed are on the input side. The agent will often read a million tokens just to patch one line of code.

I think if you are seeing something closer to 1:1 or more on the output side, there is either a problem with the agent or the codebase is new / empty.

You can infer a developer's position on AI with 100% accuracy if you ask them a related question about the customer and how much overtime they'd be willing to pull to meet a deadline for one of their internal projects. It's the same question, just worded a lot differently.

The general form is probably something like "are you willing to sublimate your ego in order to care for the parties who ultimately justify your compensation/career?"

I am currently observing AI authored tests creating a massive sense of complacency because a human no longer owns responsibility for the test suite. It's too easy to reject ownership by way of the various agent prompting schemes. I find myself enjoying the idea of it too, primarily because adding tests to even the most trivial functionality is mandatory due to the TDD policy.

Developing good tests is like an artform. Total coverage is a terrible objective. Correctness does not compose upward. It's a game of chasing ghosts if you think you can build a perfectly clean system bottom up and then magically meet the customer at the top. They're gonna kick your jenga tower over on day one.

Inappropriate aliasing of integer keys allows for silent errors in queries because it will actually return some result a lot of the time. A UUID is immune to this problem. The model recognizes its mistake a lot more reliably when previously non-empty tables start showing up empty after attempting a join.

Another "oh shit" moment was when I realized I can leave the system prompt entirely null. A properly organized agent can find its way into tool docs and iteratively work through an understanding of the environment relative to the user's prompt. The tools being more important than the prompt has actually been a massive relief for me. Magical string literals are so odious.