Matthew Halliday

VP of Product @AltimateInc. Co-founder of @incorta. Passion for design, UX, data, stories, and making beautiful things. #Analytics

[Matthew Halliday @layereddelay]

Latest @dbt_labs #ADEBench Leaderboard

#1 OSS altimate-code: 74.4% pass with Sonnet 4.6
#2 @Snowflake Cortex: 65% pass with Opus 4.6
#3 @claudeai Code: 58% pass with Opus 4.6

Try it out: https://t.co/SmtkhaJAOe

#DataEngineering #Pipelines #ADEBench #ADE

[Matthew Halliday @layereddelay]

Data Engineers have a new job. I explain where I think the future lies for Data Engineering. Agree / Disagree?

#ADE #AI #Data

https://t.co/Nj6IJEfeC3

[Matthew Halliday @layereddelay]

@KaiXCreator Historical reasons? It was better. Now? Status.

[Matthew Halliday @layereddelay]

@Palak3312 Sonnet 4.6 with a better harness

[Matthew Halliday @layereddelay]

@Venkydotdev Talk is the code.

[Matthew Halliday @layereddelay]

@DanielSmidstrup altimate-code

[Matthew Halliday @layereddelay]

@piyush784066 6

[Matthew Halliday @layereddelay]

@KaiXCreator ... Legislation

[Matthew Halliday @layereddelay]

@pmitu And your docs are the PRD for Claude?

[Matthew Halliday @layereddelay]

@Aditya_181105 It's more the harness you use than the model. I have seen better results with the right harness with "lesser" models. Check out https://t.co/SmtkhaJAOe for an Open Source example which makes this clear.

[Matthew Halliday @layereddelay]

If only people would read this blog they have a completely different vision of how to build with AI. Too many people write it off as they don't understand what needs to surround an LLM. This is the best article I've read this year on this topic. #AI #Harness
https://t.co/2M4i3Fy5qE

[Matthew Halliday @layereddelay]

@WizzOnChain_ @MelkeyDev The point we are making is the harness is more important than the model. We intentional use the older models to showcase that. We are finsihing up using Deepseek at a fraction of the cost of the models and still scoring in the top 3.

[Matthew Halliday @layereddelay]

@garyvee Could not agree more.

[Matthew Halliday @layereddelay]

@akshay_pachaar 👏
100%
That's why we dedicated our team to building a data engineering harness using the older models to win the #1 spot on the top 2 agentic data engineering benchmarks. The harness matters more than the model. Check out https://t.co/SmtkhaJAOe

[Matthew Halliday @layereddelay]

@elonmusk I heard they are going to ban altimate-code. https://t.co/SmtkhaJAOe

[Matthew Halliday @layereddelay]

Apparently it only takes one tweet for @pmarca to follow you and one tweet to unfollow you.

I wonder if he would follow you...
.
.
.
.
.
.
and unfollow before finishing reading that same tweet?

¯\_(ツ)_/¯

[Matthew Halliday @layereddelay]

@briansolis @wapo I think we shall see a resurgence in the desire for uniquely raw human experiences. The polish will just feel too much. In some ways youtube provided that back in 2010 when people were doing covers of songs and they were getting more playtime than the originals!

[Matthew Halliday @layereddelay]

@TerenceLeungSF @ChuckDBrooks @Nicochan33 @NancySinatra @alvinfoo @Hana_ElSayyed @MikeQuindazzi @Shi4Tech @antgrasso @PawlowskiMario @sallyeaves @NevilleGaunt @Eli_Krumova @baski_LA The noise would drive me nuts.

[Matthew Halliday @layereddelay]

How are you all using #claude -p along with worktrees to handle 8 claude sessions building simultaneously? Tell me your best tips. Also how are you going to handle June 15th's change in tokens with -p not being counted against your max plan?

#Anthropic #Parallelism