Building better AI tools

The very premise of the article is that tasks are needed for humans to learn and maintain skills. Learning should happen independently, it is a tautological argument that since human wont learn with agents which can do more, we should not have agents which can do more. While this is a broad and complex topic (i will share a longer blog that i am yet to fully write), I think people underestimate the cognitive load it takes to go to the higher level pattern and hence learning should happen not on the task but before the task.

We are in the middle of peer vs pair sort of abstraction. Is the peer reliable enough to be delegated the task? If not, the pair design pattern should be complementary to human skill set. I sensed the frustration with ai agents came from being not fully reliable. That means a human in the loop is absolutely needed, and if there is a human, dont have ai being good at what human can do, instead it be good assistant by doing things human would need. I agree on that part, though if reliability is ironed out, for most of my tasks, i am happy ai can do the whole thing. Other frustrations stem from memory or lack of(in research), hallucinations and overconfidence, lack of situational awareness (somehow situational awareness is what agents market themselves on). If these are fixed, treating agents as a pair vs treating agents as a peer might tilt more towards the peer side.

I’m going to but in here.

I think the point the article is making is to observe the patterns humans (hopefully good ones) follow to resolve issues and build paths to make that quicker.

So at first the AI does almost nothing, it observes that in general the human will search for specific logs. If it observes that behaviour enough it then, on its own or through a ticket, builds a Ui flow that enables that behaviour. So now it doesn’t search the log but offers a button to search the log with some prefilled params.

The human likely wanted to perform that action and it has now become easier.

This reinforces good behaviour if you don’t know the steps usually followed and doesn’t pigeonhole someone into an action plan if it is unrelated.

Is this much much harder, yes it is than just building an agent that does X. But it’s a significantly better tool because it doesn’t have humans lose the ability to reason about the process. It just makes them more efficient.

I totally disagree.

Importing an external library into your code is like using a player piano.

Heck, writing in a language you didn't personally invent is like using a player piano.

Using AI doesn't make someone "not a programmer" in any new way that hasn't already been goalpost-moved around before.

We're just disagreeing and hashing this out, but, no, I don't think that's accurate. AI tools don't watch what human operators in a specific infrastructure do and then try to replicate them. They do things autonomously based on their own voluminous training information, and those things include lots of steps that humans are unlikely to take, and that are useful.

One intuitive way to think about this is that any human operator is prepared to bring a subset of investigative approaches to bear on a problem; they've had exposure to a tiny subset of all problems. Meanwhile, agents have exposure to a vast corpus of diagnostic case studies.

Further, agents can quickly operate on higher-order information: a human attempting to run down an anomaly first has to think about where to look for the anomaly, and then decide to do further investigation based on it. An AI agent can issue tool calls in parallel and quickly digest lots of information, spotting anomalies without any real intentionality or deliberation, which then get fed back into context where they're reasoned about naturally as if they were axioms available at the beginning of the incident.

As a simple example: you've got a corrupted DeviceMapper volume somewhere, you're on the host with it, all you know is you're seeing dmesg errors about it; you just dump a bunch of lvs/dmsetup output into a chat window. 5-10 seconds later the LLM is cross referencing lines and noticing block sizes aren't matching up. It just automatically (though lossily) spots stuff like this, in ways humans can't.

It's important to keep perspective: the value add here is that AI tools can quickly, by taking active diagnostic steps, surface several hypotheses about the cause of an incident. I'm not claiming they one-shot incidents, or that their hypotheses all tend to be good. Rather, it's just that if you're a skilled operator, having a menu of instantly generated hypotheses to start from, diligently documented, is well worth whatever the token cost is to generate it.

Along the same lines, its probably little more than that. When it comes to software development, every iteration of execution/design is supposedly either faster or better based on the prior learnings for things that you have done by urself or observed very carefully.