Given how advanced these models are, I'm wondering if this is intentional. Is AI saying it has determined that strict REST isn’t a practical standard all around? Or is it just a byproduct of token efficiency or....?

I know I can steer the output with better prompting, but I'm curious whether there's a real underlying reason for this almost all-the-time output.

Comments URL: https://news.ycombinator.com/item?id=47939342

Points: 17

# Comments: 5

https://x.com/Khaliqgant/status/2019124627860050109

Here he mentions "Store trajectories " is a tent pole to see vectors of work over time which goes back to the point above. But also, will that trajectory be stored in the agent/framweork, or will it also need live elsewhere for us humans?

Also, some say they use the tickets themselves for the prompt and have Claude CLI (or alike) just work off the tickets directly.

...on the other hand... since we still have humans using the features and interacting with them, knowing what is going on and why it made a decisions (for better or worst) doesn't seem like something to let go of.

Now with coding agents, in a sense, we can read the “mind” of the system that helped build the feature. Why did it do what it did, what are the gotchas, any follow up actions items.

Today I decided to paste my prompts and agent interactions into Linear issues instead of writing traditional notes. It felt clunky, but stopped and thought "is this valuable?" It's the closest thing to a record of why a feature ended up the way it did.

So I'm wondering:

- Is anyone intentionally treating agent prompts, traces, or plans as a new form of documentation? - Are there tools that automatically capture and organize this into something more useful than raw logs? - Is this just more noise and not useful with agentic dev?

It feels like there's a new documentation pattern emerging around agent-native development, but I haven't seen it clearly defined or productized yet. Curious how others are approaching this.

Comments URL: https://news.ycombinator.com/item?id=46882676

Points: 3

# Comments: 4

I’ve seen this before with non-AI development. Usually it comes down to skill. People haven’t yet learned how to decompose a big problem into smaller, concrete steps. With some guidance, they improve. We go from two big "8s", and find out we can release a few 2s and 3s of value over time.

But with AI system development, I’m not sure if this is the same issue or if the nature of the work really is different. The engineers argue that it’s harder to "shrink" AI work into predictable, incremental pieces because outcomes are probabilistic, not deterministic. And that we can't just break them up since they rely on one another contextually.

So I’m curious:

- Are others experiencing this shift?

- Is this just a new version of the same problem-decomposition skill?

- Or, is building AI systems genuinely a different game we have to calibrate expectations around?

- And either way, how have you adapted your process to deal with these longer, less predictable, larger tickets?

Would love to hear what’s working for people.

Comments URL: https://news.ycombinator.com/item?id=46591688

Points: 1

# Comments: 0

Ironically, now that agents are genuinely good at coding, I find myself “writing” more shell scripts.

They’re robust, portable, and work in almost any environment I care about. Agents are very comfortable producing them, they execute fast, and they’re great for automation.

What I’ve ended up with is a growing set of automations built around simple, interactive shell script menus that handle things I do all the time. When I want to improve one, add branching logic, or tweak behavior, I just ask AI to mod the script and move on. I kind of love it.

It feels like a shell script revival, at least for me.

Curious if others are experiencing the same shift or if I fell into some weird bubble/anti-pattern.

Comments URL: https://news.ycombinator.com/item?id=46260216

Points: 1

# Comments: 5

AI looks a lot more like the former. Some companies will fail, valuations will swing, but the underlying technology isn't going anywhere. In fact, many of the AI firms that will end up mattering are probably still undervalued because we're early in what will likely be another decade long technology expansion.

If you're managing a portfolio that needs quick returns and can't tolerate a correction, then sure, it probably feels like a bubble, because at some point people will take profits and the market will reset.

But if you're an entrepreneur or a long-term builder, that framing is almost irrelevant. This is where the next wave of value gets created. It's never smooth and it's never easy, but the long-term opportunity is enormous.

For example, we're planning to host monthly events in SF to help other founders and operators learn something genuinely useful. (not "use our product," but "here's how to do x,y,z, sponsored by us.") Then we follow it with a: "Join our Discord to keep the conversation going and meet people like you."

We're still small, so we hired someone entry level to post content on social every day to build momentum, with the hope of identifying a spark we can double down on messaging-wise.

I could go on, but if no one reads or comments, it disappears into the ether. So I'll stop here and hope this grows into something we can all learn from, without fluff or shameless self-promotion getting in the way.

Comments URL: https://news.ycombinator.com/item?id=46222009

Points: 1

# Comments: 0