JP Hwang

Trying to read a book

[JP Hwang @_jphwang]

Hi there, my name is JP. I’m a developer educator.

I create:
- courses https://t.co/FKWrK2qDsR (@datacamp), https://t.co/wqQKpqe316 (@linkedin learning)
- educational diagrams (https://t.co/BCr1h2tzyL) and projects (https://t.co/wxrj6Uu2UD)
- blogs & videos.

Let’s connect ❤️

[Zhizhuo (Z) Zhou @zhizdev]

Why is AI slop so frustrating to read?

AI writing often obscures main ideas and invents awkward phrases, making sentences more complex than necessary.

Many of these slop patterns violate writing principles established in English writing books.

In our benchmark, AI writing breaks these principles far more often than human writing, with Claude Code replies by far the worst offenders.

We show examples of the most common slop patterns in this thread

[JP Hwang @_jphwang]

@walterra Controversial message but I can get behind it.

[JP Hwang @_jphwang]

@CShorten30 @bobvanluijt Congrats @CShorten30 that's some achievement! 🎉

[JP Hwang @_jphwang]

I'd love to see your favourite developer experience (and agent experience) resources, actually.

[JP Hwang @_jphwang]

A preview of my blog post for @weaviate_io about DX in the agentic age.

[JP Hwang @_jphwang]

I'm occasionally blown away at how good agents are at some things now. Here's an output from a throwaway, "Oh, also - get a subagent to make a chart"

[Não Intendo @blognaointendo]

É a primeira vez que eu vejo uma música eletrônica analógica

[JP Hwang @_jphwang]

I can’t recommend this talk enough by @richcampbell

What a great storyteller and writer.

https://t.co/sIVmrlRF3x

[JP Hwang @_jphwang]

Super interesting to see - although obviously this is *only* based on open models.

Jarring to see that OpenCode is so low on this list - *consistently* low in fact for every model used here.

[Joël Niklaus @joelniklaus]

I burned a shitload of compute so you don't have to. Releasing the Harness Leaderboard.

Which harness should you run your open model in? I ran 28,849 trials to find out.

The leaderboard compares nine coding-agent harnesses on open models. In each comparison only the harness changes: the model, provider and reasoning effort stay fixed, and every task gets three attempts.

The trials ran GLM-5.3-Flash, DeepSeek-V4.1-Flash and Flint on Terminal-Bench 2.1 and DeepSWE, for $4,746 of inference.

Averaged over the three models, mini-swe-agent (@KLieret), a bash-only agent, solves 51.4% of tasks and @openclaw 27.3%. Four others are statistically tied with mini-swe-agent.

One of them is Pi (@pidotdev), at 49.4% for $0.29 per solved task, about 45% less than mini-swe-agent's $0.53.

The best harness depends on the model: mini-swe-agent on GLM-5.3-Flash, Hermes Agent (@NousResearch) on DeepSeek-V4.1-Flash and Kimi Code (@Kimi_Moonshot) on Flint.

@opencode costs $3.35 per solved task, 11.7 times as much as Pi.

See Harness Hub's Harness Lens for what the nine harnesses do differently. It documents the prompts, tools and compaction of 34 coding-agent harnesses and links every claim to its source.

Leaderboard: https://t.co/AiWiQtBOz5
Harness Lens: https://t.co/I8jbcWa9JH

[JP Hwang @_jphwang]

Cries in swarf

[Carl Schoeller @CarlSchoeller]

Getting parts made in China is so cheap and fast I’ve started redesigning everyday objects to be custom made.

Latest is a teapot 5-axis machined from billet stock (with corresponding teacup)

[JP Hwang @_jphwang]

@mufabal_vettori Lol

[JP Hwang @_jphwang]

Hate-reading my way through AGI Chronicles because it’s choc full of insane details like these.

[JP Hwang @_jphwang]

@michielmv @weaviate_io More stuff coming on that too :)

[JP Hwang @_jphwang]

Agent experience (AX) updates for
@weaviate_io:
- For Markdown docs, add `.md` to the url
- Agents sending `Accept: text/markdown` are auto redirected
- Readability-based fetchers correctly get all code tabs.

Lots more coming.

@ me if any Weaviate + agent issues

[JP Hwang @_jphwang]

@SidhantGoyal8 @weaviate_io If we have set up the build correctly, it should pick up the right snippets as they evolve. :)

[JP Hwang @_jphwang]

@kevinroose Thank you for your journalism!

Boggles my mind reading this so I cannot imagine what it's like being there in amongst this stuff in SF.

[JP Hwang @_jphwang]

@BrettDavies @weaviate_io Have created a PR here: https://t.co/0BdvarqZh1

Thanks for the tip @BrettDavies and god luck with https://t.co/51fWU2CPhG :)

[JP Hwang @_jphwang]

@BrettDavies @weaviate_io Thanks very much! I will look into it :)

[JP Hwang @_jphwang]

@weaviate_io Oh lol I should note this relates to our docs 🤦🏻.

[JP Hwang @_jphwang]

Interesting finding:

In our experiments agents would often pick up the "nearText" example in the Quickstart and use it almost like best practice.

But adding a little admonition like this is a cheap way to steer them while enabling human learning.