Ryan D. Hatch

AI Privacy. Product @ Hydra. Family man. Saved by Grace.

[Ryan D. Hatch @rdkhatch]

People won't buy unless they visualize life with your product + how it changes the way they live/work. cc: @zgohr #jtbd #prodmgmt #startups

[Ryan D. Hatch @rdkhatch]

@DanielMiessler This is exactly the right direction. I'm using graphs now now without any supportive tooling, and it's a game changer. Looking forward to what's coming.

[Ryan D. Hatch @rdkhatch]

Stord is bringing Amazon-style delivery to all brands.

[Sean Henry @seanhenry]

Ten years ago I started Stord because nobody had built the infrastructure for what happens after a consumer clicks buy.

Today we raised $250M at a $3 billion valuation.

Commerce now has the physical intelligence layer that merchants deserve.

Full piece: https://t.co/DhDur7veoh

[Ryan D. Hatch @rdkhatch]

@StiofanPhadraig @libsoftiktok Sounds like a good case.

[Ryan D. Hatch @rdkhatch]

Elon determined to appeal ruling in favor of OpenAI, which cited statute of limitations

https://t.co/fh2OHd8pzP

[Ryan D. Hatch @rdkhatch]

@grok Why is this?

[Ryan D. Hatch @rdkhatch]

If you think AI is superintelligence... you are mistaken.

SOTA 2026 is sadly not able to make basic decisions.

Car Wash question - even with prompt assistance.

Results:
------------------
GPT 5.5 - Winner
Opus 4.7 - Lose
Grok - Lose

[Ryan D. Hatch @rdkhatch]

@SeedsForbidden Honestly amazing. Inspiring. And what great parents you had to let you chase your dream. Amazing what intense curiosity + boldness can accomplish.

Your secret seemed to be the network you built along the way.

Thank you for sharing!

[Ryan D. Hatch @rdkhatch]

@lennysan Still hype more than real. Directionally correct though.

[Ryan D. Hatch @rdkhatch]

@HarisEbrat @BrianRoemmele The answer is yes. Biology runs on software.

[Ryan D. Hatch @rdkhatch]

If you're in Horizontal SaaS... You should worry.

If you're in Vertical SaaS.... Great place to be right now.

Stay close to customer outcomes.
Solving problems > Shipping capabilities.

[Ryan D. Hatch @rdkhatch]

@sudoingX Very cool. Q: What's your process to arrive at such a detailed build prompt?

[Ryan D. Hatch @rdkhatch]

@0xSero Would enjoy connecting for a drink in SF this week, if you'll be in town. Several AI eng parties scheduled on Luma.

[Ryan D. Hatch @rdkhatch]

@0xSero Are you going to be @ AI Council conference this week?

[Ryan D. Hatch @rdkhatch]

Agent Skills = Risk.

And the solution is likely 3-fold:

1.) Skill Trust
This is the root thing that needs to be addressed. Skill scanner, or even better - certified reviews by trusted parties.

2.) Sandbox control
Harness control over agent sandboxes: what data goes in / what data goes out / skill access, etc.

3.) Smaller Skills
More Determinism, less LLM. code that can be scanned / checked using traditional tools + much more cost effective. Easier to scan skills, smaller risk surface area.

[Ryan D. Hatch @rdkhatch]

Anyone else find it funny that you start to view real people as AI agents. We are prompting each other. Each of us with limited attention / context window.

[Ryan D. Hatch @rdkhatch]

Game Changer, if their model is good.

12M Context, Lower Cost.

"Much of standard attention's compute is models talking to themselves about words that don't matter."

[Alexander Whedon @alex_whedon]

Introducing SubQ - a major breakthrough in LLM intelligence.

It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),

And the first frontier model with a 12 million token context window which is:

- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus

Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).

Only a small fraction actually matter.

@subquadratic finds and focuses only on the ones that do.

That's nearly 1,000x less compute and a new way for LLMs to scale.

[Ryan D. Hatch @rdkhatch]

@pupposandro It's beautiful. Be proud of that amazing mess.

[Ryan D. Hatch @rdkhatch]

@rawsalerts Sounds like a developer made a mistake & blaming the model.

Although... deleting backups? That sounds intentional.

[Ryan D. Hatch @rdkhatch]

Yes, we need many more US-based Open Source models.

However, you can run Deep Seek 4 and others on your own private Infra using @get_hydrahost.

Private Inference. Totally secure. There's a LOT more infra being built outside of Anthropic/OpenAI. And that's great for the open source community.

[Ryan D. Hatch @rdkhatch]

I'm excited to try this.

We need multi-agent orchestration, customizable workflows flowing upstream to downstream, parallel operations, model selection for cost efficiency.

Harness is a big deal.

[Rafael Aleksandryan @NoVibeJustOrch]

So, don't take my opinion on LaunchApp Animus CLI.

This is what grok says, if that sounds like something you want repo in my profile. If you want to get stuck in the AI slow lane, then do whatever it is you're doing.