Aaron Daub

CEO, Founder - Labyrinthine Software

We're pushing the pareto frontier of Coding Agents from Toronto, Canada

[Aaron Daub @aaron_daub]

@bygregorr @thebuggeddev how much does this bug you and would you want it to be fixed forever?

[Aaron Daub @aaron_daub]

Because the best software is always simple.

And the core task of designing good software systems is keeping it as simple as it can be. No simpler.

[Aaron Daub @aaron_daub]

You need as model as big as your input. We have models so huge your input can be anything, right?

If you need simpler output, you can use smaller models. The good news is: the world needs simpler software.

[Aaron Daub @aaron_daub]

@provisionalidea What should we know to ensure our interfaces encourage engagement, focus, and flow?

[Aaron Daub @aaron_daub]

@zachmoskow @concentrateai I want to know more about how you guys think of evals :)

[Aaron Daub @aaron_daub]

We can make iOS apps faster than anyone else. This morning, we were 10x faster. After some fable time, we're 33x faster.

[carried_no_interest @carrynointerest]

@JayaGup10 Utility per token will rule all. Both labs have this problem and it is currently unsolved for both labs. The lab that figures it out first will own enterprise:

https://t.co/pKS5QBUIdR

[Aaron Daub @aaron_daub]

@carrynointerest @JayaGup10 Labyrinthine is the only Lab focused exclusively on this.

[Aaron Daub @aaron_daub]

@gakonst The large majority of software engineering tasks can be handled by Gemini 3.1 Flash Lite

[Aaron Daub @aaron_daub]

@garrytan Software engineering is the reification of intent. Don't conflate calcification for reification.

[Aaron Daub @aaron_daub]

It's a shame coding agents are powered by models so strong, U.S. Export Controls step in.

If only there was a coding agent that could do more with less. Stay tuned.

[Aaron Daub @aaron_daub]

@MartinGTobias #1 NPS for all b2c coding agents, EOY

[Aaron Daub @aaron_daub]

@cantrell @reve The output quality looks amazing! How does the intermediate representation affect the quantitative performance at inference time?

[Aaron Daub @aaron_daub]

@backseats_eth Hi, it’s me, Companies. We have been doing the benchmarking.

It is cheaper, by a lot!

[Aaron Daub @aaron_daub]

@carrynointerest Using the Big Models to serve user traffic is overkill. Instead, use Big Models to make things simpler for Small Models.