Meiring de Wet

Building payments and membership software for creators.

[Meiring de Wet @meiringdw]

https://t.co/VfzPpmJvQE

[Meiring de Wet @meiringdw]

https://t.co/6DriTWUUdg

[Meiring de Wet @meiringdw]

@czue just signed up, this looks awesome!

[Meiring de Wet @meiringdw]

@jonathan_wilke Your agents need to understand it though, without it they'll make even a bigger mess

[Meiring de Wet @meiringdw]

@AdamRackis it's an incredibly satisfying feeling to code well by hand

[Meiring de Wet @meiringdw]

@rtbenfield i love it, there are some quirks if you have an accent, but I can even steamroll through and LLMs will make sense of it

[Meiring de Wet @meiringdw]

@housecor the process shouldn't be any different than it used to be if you're serious about maintaining quality software and products. If you don't review the code, any future work (agents or otherwise) will assume the standard that's been set. Absolutely review code, always.

[Meiring de Wet @meiringdw]

@arvidkahl Something like https://t.co/EKpR0t6J2o for agents

[Meiring de Wet @meiringdw]

@tdinh_me https://t.co/WhVmkKpDlO and https://t.co/AnXNvgPdn9

[Meiring de Wet @meiringdw]

@allgarbled Pleeease share the skill!

[Meiring de Wet @meiringdw]

@boringmarketer There should be a FOCUS step in there somewhere to keep the guy from wandering off topic

[Meiring de Wet @meiringdw]

@AKBrews necessity breeds innovation :)

[Meiring de Wet @meiringdw]

@soohoonchoi @anuraggoel @hatchet_dev Also started using hatchet to replace temporal. easier to set up amd self host

[Meiring de Wet @meiringdw]

@boringmarketer Pasted this directly into CC.."the audit now applies to all future landing-page work in this repo without you having to ask"

[Meiring de Wet @meiringdw]

@MrPunyapal No. Agents like to do it though

[Meiring de Wet @meiringdw]

@brankopetric00 what did you use for structured logs instead?

[Meiring de Wet @meiringdw]

i edited code by hand today 🥳🫠

[Meiring de Wet @meiringdw]

@ElitzaVasileva good luck!

[Meiring de Wet @meiringdw]

@saen_dev @jxnlco Definitely. crucial step for support agents as well

[Meiring de Wet @meiringdw]

Spent a week optimizing a RAG pipeline for a support chat agent after reading @jxnlco's RAG series.

recall@1 went from 23% to 50%.

This is just 1 of 6 RAG evals (only retrieval quality C|Q). Faithfulness and answer relevance are still untested.

Plenty of room left to improve.

[Meiring de Wet @meiringdw]

@mattpocockuk How would you approach an "analyze" type skill that is inherently multi-step? Like, fetch data, check for patterns based on some rules, give conclusion