event

AI Agent Updates

25 items · started 2026-09-19 · closed 2026-10-04

  1. King Harald V of Norway died on 28 August. I wanted to know which models actually decide to use a search tool when the answer depends on something after their training cutoff, so I wrote a small harness.

  2. Author here. This is an open-source (MIT) experiment, not a product.

  3. I can't imagine it's just me. In recent weeks/months, Claude (Sonnet 5) has picked up this really ugly habit of insisting on disagreeing, or "gently pushing back" (UGH, it keeps and keeps and keeps using that phrasing), on at minimum one t…

  4. Got a new toy to play with in powershell now

  5. Saw a topic the other day about making a demo reel. Tried it on sonnet 5 out of curiosity.

  6. I've heard that opus 5.5 is the best model so far, but every time I use it, it just spawns sonnet 5 agents for everything, is it really worth saving usage by using sonnet 5 subagents? Should I tell it to just spawns opus 5.5 agents with Me…

  7. I made an envelope budgeting app inspired by YNAB and Actual Budget. The original idea was to build something just for personal use.

  8. EDIT: Just to make it clear, because I think people didn't understand my post, because English isn't my first language. What I need to know if is worth to pay money to use Claude?

  9. I know system prompts have been published, but I didn't see Sonnet 5 anywhere and I thought this behavior was interesting either way. I'm working with sonnet 5 on an a guardrails/sandbox pi-agent extension in the terminal, and I had Claude…

  10. I'm coming up for air from the bottomless evals ocean. I built a Claude Code plugin for writing code that uses Jev.

  11. I run coding agents in waves: a lead splits the work, implementer agents write the code, and a separate reviewer agent must pass every PR. I priced five days of transcripts at API list prices.

  12. Hi guys I use sonnet 5 on medium or high basically all the time, I am scared to use haiku because what if the task isn’t being done as good and Im scared to use opus because of the credit usage. What is a good way to use all 3 models in a…

  13. Like the title says, what am I going to be missing out on or what will be lacking if I am limited to only sonnet 5. 90% of the work is corporate c# and js work.

  14. Been using 5.5 since morning on GHCP. It feels like magic.

  15. Jev vs. LLMs on Reddit's "Am I the Asshole?" Bottom line: In the original seven-setup benchmark, Jev came second, behind Sonnet 5 (Brier score 0.369 vs.

  16. I have a pretty straightforward workflow - batches of work and a plan around what depends on what and what can run in parallel on worktrees. Implementer, Reviewer, Fixer.

  17. I regularly use Opus 4.6 and Sonnet 5 for coding a Next Js application. Today I implemented a new large feature with Opus 5.5.

  18. Data from artificialanalysis.ai It's honestly shocking how competent Luna is at higher reasoning levels while being incredibly cheap, even if it's a bit slower. While GPT-6 Sol costs ridiculously less than 5.6 Sol while being more compente…

  19. TL;DR: Three programming assignments, from easy to hard. Every model got the exact same instructions, one attempt, no help.

  20. Sonnet 5 nearly drove me insane this weekend, we all know why. I eventually settled on the instruction to always begin answering a direct question with the bottom line up front.

  21. Estuve trabajando hoy en la mañana en claude code con Opus 4.8 y estaba trabajando perfectamente, pero cuando llego la notificacion de la nueva version, tal vez me estare imaginando cosas, pero empezo a olvidar cosas con un contexto de 225…

  22. So i've been running the same little test on the latest Sonnet, Opus, Astra and Fable models. The prompt isn't the typical pelican bicycle test, but basically I asked these models to make a svg of a lighthouse on a rocky headland at night,…

  23. I run Claude Code and a few other AI agents for hours a day on real work. Here are six specific things that eat time without warning, what actually happens, and the exact fix.

  24. Its so weird that Claude (sonnet 5) manages to explain exactly why it is wrong, and then back up its wrong answer. I feel like Claude would have been trained to understand this, is the image to confusing for it?

← all threads