verdverm

it occurs to me how much better these endeavors might go if the prompt was

"write a program to transpile A to B" instead of "port A to B"

There are a lot of React Native apps out there, the issues I see are more around platform glitchiness or workarounds than performance.

The main thing to watch out for is blocking the rendering or event loop. I've definitely seen apps where a slow API call blocks everything else. So less about performance and more about knowing how to do concurrency well.

If you aren't doing native apps, my recommendation is to use the same language across full-stack apps. You can share things like zod types and validation logic (in the frontend for quicker user feedback, on the backend because you never trust user provided values)

The actual title is

> Introducing cf: the agentic CLI for the entire Cloudflare API

emphasis my own to highlight the key word missing in the HN title

re: wrangler, my biggest gripe was inconsistency between dev and prod

I agree that writing is much harder, largely because there is no way to get concrete feedback for "does it work"

I still think they produce shoddy or sus code too often, an artifact of the current generation's training to try anything and everything until it "completes the task". They have a hard time even with that concept, which is part of "coding" imo

consumers buy brands (at least in america) and this is the real moat of Anthropic, OpenAi, Google, and Meta

Just rolled out our in-house review agent, catches things the paid offerings do not, but comes with a comment wall... we're rolling with it anyway since it is catching real defects.

We still require a lead/cto to sign off, one cannot merge their own PR, they remain tied to a human at creation time, so there are at least two pairs of biological eyes on every PR.

...tbd what the cognitive effects will be for us.

Kudos to cloudflare on this, it looks like a great CLI, I especially like the `cf cli search`.

Related, Matt Pocock's skill for having agents generate an interactive bash script for things only humans can do (or should do), presented in the context of devops like activities

https://github.com/mattpocock/skills/blob/main/skills/produc...

I have found that having the agents write scripts to use tools like this is better (less tokens, more reliability, fewer side quests) than giving it to them directly with markdown they may or may not follow on any given day.

The other benefit to this is that you can put scripts on either side of the agent and remove all credentials from their process, removing whole classes of issues you don't want to have to tell your boss about. CLIs like this are great for read-only debug sessions, but if you are going to make modifications to your cloud infra, keep doing IaC.

Depends on the calculation and context, it is definitely advisable to use tools for complex or high-risk situations, dividing the dinner bill by 2 or 3, that should not require a calculator or one of the thousands of bill splitting apps, perhaps it is a bit harder when you want divide bill by items, but not a high risk situation.

practice makes perfect, exercise your brain like your body

> no longer are concerned with writing or editing literal code

I'm not typing keys, but I am very much still concerned about the quality and nature of the code. Coding to me is more than pushing keys

> if you can define a really good PRD and sophisticated technical specs

I still believe we cannot waterfall software, the idea seems like taking a step backwards. How often do we learn about an unforeseen complexity only after getting into the implementation?

In my experience with agents, it's better to be iterative and in-the-loop. Start with a decent description, have them research the code/issue, write up an initial plan/design, work iteratively on writing code and updating design doc, review and finalize the code and markdown. Then future agents will have some resources to shortcut understanding the code base.

What is coding? Pushing keys or knowing the right keys to push?

Ai can push a lot of keys very fast, but not always the right ones

if they need to be reminded to follow the coding standards, visible in the very code they are working on, what has been solved?

I'm well on my way down this path too, but that is a process implemented in markdown. I don't see coding as solved, or a solvable thing at all, like writing

To solve something, there must be a defined problem, what is the problem that was solved. Or perhaps it is just "coding is solved" is the turn of phrase de jour be ause we haven't yet found a more succinct and accurate way to describe the paradigm shift

When it comes to non technical people using Ai to build things on code, the outcomes are on average pretty poor, which i see as evidence that the driver and their expertise behind the Ai matters a lot. A notable example is Terence Tao's conversation with ChatGPT, us math normies could never have done that. The same applies to coding agents ime

even setting aside the artistic side of the profession, I see these things make all sorts of basic mistakes, so even the mechanical part doesn't seem "solved" ime

I had agents deep diving OpenCode source over the weekend, writing research docs, and helping me with a set of plugins. I learned quite a bit from the process and markdown walls, enough so that in later sessions I was able to point the model at where it was conflating concepts and help narrow its search.

Ai can help you learn if you are intentional about it.

When you say "solved coding," what does this mean, what does it look like?

I have strong disagreement because it sounds like, by analogy or proxy, we have also "solved writing"

The tables have normally only gone to {12}x{12} and most young children have no issue. Why would we cut back on this?

The memorization means that you can extrapolate the more complex concepts. If you don't have the very basics, what are you even going to extrapolate on?

There are basic skills that everyone should learn, being able to do simple math in your head is a building block for more complex reasoning.

I once had a roommate who I encouraged to divide our grocery bill by 2 in his head for splitting the bill, by the end of the year he was quick and accurate.

the worst thing I heard him say is that he doesn't think kids should learn their multiplication tables anymore on the Ezra podcast

while I generally agree on his open weight stances, I lost all respect in that moment, everyone of these people are so out of touch

they could earn some good will by making all of the models and tools open source, set the "pacing" for Big Ai labs in the US

interesting analysis, though I disagree that it's a harness moat

if you look at OpenCode, you can see their awareness of this stuff, with conditionals and custom prompts/tools depending on the current model, notably an apply_patch for gpt models where the rest get the normal edit tool

Sandy Monroe would likely agree, he tears down cars and sells what he learns, all the big auto companies buy info about competitor cars

Why should Ai be different?

Fortunately open weights are likely to dominate and it becomes a permissionless ecosystem

people have forgotten how useful deterministic code generation is (and will remain)

When the Go team ported the original compiler from C to Go, they wrote a program that did ~99% of the work

https://www.youtube.com/watch?v=QIE5nV5fDwA

https://go.dev/talks/2014/c2go.slide#18

sounds like 2x the code that no one understands, one more reason to never consider using copilot again

would be curious to know how many times "unsafe" appears in there, have seen rust devs comment on how the ais like to use unsafe to work around difficulties with memory management, like how they will sometimes subvert tests

I think the entire framing is wrong, I don't see coding as a "problem" which can be "solved," sounds the same as "we solved writing," like what does that even mean or look like?

lord of the flies, modern capitalism version?

what happened within the agents does not matter

what we would like to see is your settings and configuration, maybe the task(s) you gave them and any code/scripts around them, where did they run from (your laptop vs cloud vm)

why did they even have credentialed access to change credit cards? Sounds like you didn't do the basics for isolation

the agents go on side quests, all the time... super frustrating, but I suspect between that tendency and asking about a UI (image), you racked up a bill with legitimate requests

There is a reason some of us preach "stay in the loop" and I hope you now understand why we do

user errors can happen before you write the first prompt or hit enter to start the hacking test, eg. not even doing the basics for isolation / airgap

> doesn't imply that they're competitive with frontier models for the most advanced tasks

this is different and nuanced from the "not even close" or "they are trash" that the other person in this thread has opined, note how they also claim Claude is way ahead of OAI as well