Shred

scaling

[Shred @_shredr]

rule zero for scaling with multiple agents: what can be done orthogonally

@grok amdahl's law

[Shred @_shredr]

@HansonZachary @BrianRoemmele but Claude told me it didn’t

[Lisan al Gaib @scaling01]

Cold take on what comes next:
- OpenAI will flourish
- Anthropic will continue to be profitable
- Google will not catch up to Anthropic or OpenAI
- no chinese company will catch up to Anthropic or OpenAI
- the highest tier of intelligence will become a luxury product that only companies and multi-millionaires/billionaires can afford
- most of the companies that invested massively in them will have massive returns
- SpaceX’s AI will be fine and on par with Google by end of year
- Nvidia will become the first 10T company

[Shred @_shredr]

2026 is scaling swarms by optimising model selection and delegation

Next layer on task specific models is delegation/orchestration specific models which optimise model choice

Swarms will have higher productivity token credit minutes than single agents

[Ashutosh Maheshwari @asmah2107]

Prediction : By 2026 end, the entire industry might switch.

We’ll likely stop spending money on frontier models for plumbing jobs.

Instead we’ll use cheap, fast, task-specific models for the boring stuff and save the expensive frontier models only for the hard reasoning parts.

Result?

Agent costs drop dramatically (5-10× cheaper in many cases), agents feel snappier, and small teams can actually compete.

[Shred @_shredr]

the agent bottleneck is world modelling
model intelligence is not

Memory is just a subset of world modelling, L1 substrate

2026 is for building out the L2 world models

[Shred @_shredr]

forget harnesses, build foundation agents

[Tom Dörr @tom_doerr]

Curated papers mapping foundation agent cognition

https://t.co/PHnFVfojEw

[Shred @_shredr]

@kaledora @AviFelman convergent evolution on new a new substrate

1. Speedrun everything that already existed
2. Expose mechanisms only possible on the new tech
3. Replace old substrate

[Shred @_shredr]

@milesdeutscher Curious how you objectively evaluate whether the cheaper model performs effectively enough for brainstorming/planning?

I find I have much more success when I invest in flagship models for vision+planning+architecture

[Shred @_shredr]

@tlanderso @0xSero @OnlyTerp Eliminating pointless drudgery eliminates a large portion of the private sector and majority of the public sector

The “administrator” class

[Shred @_shredr]

@shawmakesmagic Probably because harnesses are commodity so distribution matters more than the execution

If your idea only requires paying enough credits to anthropic there’s no moat - brute force searching the solution space

[Shred @_shredr]

You are the bottleneck

[Shred @_shredr]

Time to pivot from bits to atoms

[Shred @_shredr]

@RNR_0 yeah, works cross platform too

[Shred @_shredr]

@Argona0x How did you measure rule adherence?

[Shred @_shredr]

@finn_builds @0xSero Your cross domain pattern recognition will detect it if you break down concepts to pieces you can logically validate in brain

[Shred @_shredr]

@fartface2000

[Shred @_shredr]

@idosal1 How many tokens to achieve this?

Agent swarm UIs are the only way.

[anabology @anabology]

Changed my main device to the Minimal Phone to take back my attention and avoid blue light. Took some setup but feels amazing so far.

[Shred @_shredr]

@ahboyash @printr @OmniCatCoin @LayerZero_Core 🫡