Lasse

I code and automate stuff with AI

[Lasse @lassehl92]

Google is shutting down Jules, its async coding agent. Read-only from Nov 11 and gone Dec 31. A bit ironic that the export guide tells you to use Claude Code. It doesn't sound like they even use Antigravity themselves 😂

https://t.co/E33QeZIRvq

[Lasse @lassehl92]

I don’t really see a problem here. It seems to me that the OP is just Simon’s observations, and he’s simply talking about what he observed. I don’t see how this has anything to do with being financially invested in it. He’s not trying to sell you things in this post.

My observations align with his tho. Most normies hate AI, either because they now have to pay more for their gaming hardware or because they simply don’t know how to take advantage of AI, so they’re getting defensive and angry at the people who can.

[Lasse @lassehl92]

Just because people promoting AI have a financial interest doesn’t mean the technology doesn’t work or isn’t the future. The people who invented the car did it for money too. Would you rather take a horse carriage or a car to work? I pick the car.

AI already writes better code than most people, and it can code entire production apps from scratch without you ever having to touch the code. This has been true for the last 6 months or more.

[Lasse @lassehl92]

Still looking pretty desperate to get everyone thinking the same way as you 😂 This is exactly the anger stage the original post was talking about, and popping back days later doesn’t make it any less obvious. I saw your earlier post and skipped it because it wasn’t worth engaging with.

[Lasse @lassehl92]

@KaiMeyer @SimonHoiberg Calling it fear mongering and saying he has no nuance came across pretty heated tho

[Lasse @lassehl92]

@KaiMeyer @SimonHoiberg You're kind of proving his point here. That's exactly the anger stage he circled 😂

[Lasse @lassehl92]

@charlesmendez I don't think it's dead, it's just called agentic engineering now. And it's a big jump from building stuff for fun to building an app a whole company runs on.

[Lasse @lassehl92]

@DarlingtonDev @theo I don't think Elon minds. Anthropic pays him rent, so SpaceX wins anyway.

[Lasse @lassehl92]

I suspect the reason they don't do it is cost. If everyone runs the same base OS, it's basically only the diff that takes up space, and it's a lot easier for them to maintain one image. It's also not really meant to be used like a normal desktop. The bot mostly uses CLI tools on it and only opens the browser when it has to.

[Lasse @lassehl92]

@ryanrhughes It's kinda funny how agents always get it wrong. Sometimes they estimate days or weeks for a task and then finish it in 1-2 hours.

[Lasse @lassehl92]

I have a bot that's basically my personal assistant. Every morning it sends me a brief with my calendar, mail I need to reply to and packages on the way, and it cleans up my inbox while it's at it. On the 1st it goes through my mail for bills and subscriptions and tells me if a price changed or something's about to renew, and every week it reviews my usage and optimizes it.

Then I have one for home automation and network stuff that runs a check every Saturday, one that helps me with X, a research bot, and a few project specific ones for coding etc.

[Lasse @lassehl92]

I just reserved my Cloudflare Wallet tag: https://t.co/TOgCOvF9hr.

Reserve yours now at https://t.co/dyuAIxGuCJ

[Lasse @lassehl92]

@KI_Vater I split it too. One agent writes it, a different one reviews the PR, and then one of my bots reads the diff before anything gets merged. I don't really want the same agent grading its own work.

[Lasse @lassehl92]

Anthropic gave 50 analysts a stats task with Claude Code. On calls like "measure crop yield per hectare or per unit of cost" (both are reasonable), a coin flip secretly decided which side the agent argued for. They went with the agent 86% of the time. Makes me think I should decide what I'd pick before I ask the agent, so I can tell if it actually convinced me.

https://t.co/OXCc6DHk1X

[Lasse @lassehl92]

Some people treat agents like regular code. They expect the same result every time and get disappointed when it isn't. You don't expect that from people either. You just account for it.

[Larsen Cundric @larsencc]

Rule I'd give anyone coding with agents:

Before you ask for the change, ask how you'll both know it worked.

> a test that fails right now
> a command you can run yourself
> a log or screenshot of the real thing
> "done" with no proof gets sent back

Most bad agent sessions start with a vague ask and no check.

What's your rule for trusting an agent's "done"?

[Lasse @lassehl92]

@larsencc Agreeing on the task up front helps a lot too. I make a plan with the agent first, then let it finish and read the PR after, so we don't go back and forth as much.

[Lasse @lassehl92]

@samwhoo @thorstenball Yeah. Some people treat agents like regular code and expect the same result every time, so they get disappointed when it isn't. You don't expect that from people either. Same goes for agents, you just have to account for it.

[Lasse @lassehl92]

I think a lot of "agents aren't good with our setup" comes down to habit. People are used to making every decision themselves, so they end up fighting the agent's choices instead of rolling with it. Telling it your preferences is fine, but if you constantly interrupt it, you're gonna have a bad time.

[Lasse @lassehl92]

@Stephanvs @thdxr Would you trust a product with all your signing keys though? That's kind of what he's guarding against.