Pinned Tweet
i’m starting a new startup in singapore and i’m calling it directionally local.
people who are serious about software should still make their own hardware.
directionallylocal.com
people who are serious about software should still make their own hardware.
directionallylocal.com
13
6
41
4,233
very weird example and reasoning. sounds. sounds like an orchestration problem rather than a parallelism (sub agent) problem?! @claudeai
“For example, one company found the dozen of specialist subagents within its customer-facing agent made a consistent on-brand experience difficult to maintain and added latency that hurt the user experience.
….
In these cases feedback is often that most of this structure was doing work the model no longer needed done for it, and that one clear system prompt, 4-5 thin data tools, and firm output guardrails would cover most of the use case.”
“For example, one company found the dozen of specialist subagents within its customer-facing agent made a consistent on-brand experience difficult to maintain and added latency that hurt the user experience.
….
In these cases feedback is often that most of this structure was doing work the model no longer needed done for it, and that one clear system prompt, 4-5 thin data tools, and firm output guardrails would cover most of the use case.”
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
claude.com/resources/arti…
0
1
1
143
flow state is the when you do something that aligns all your intrinsic and extrinsic motivations
0
0
2
217
it is. my self expression these days: directionallylocal.com
“software is an act of self-expression“
0
0
0
45
fascinating
It was a joy to connect with the @alibaba_cloud team in Hangzhou. Their hospitality was next level, and we're going to work together on making the Omarchy China version perfect. No VPN necessary, all the major Chinese apps preinstalled, dedicated ISO. It's going to be great!
0
0
0
79
i’ve been using my phone so much lately that my eyes have gotten weaker. just in the past month. i can feel it. dang
0
0
0
65
$700 for this in san francisco is an awesome deal
Lived in this building for 5y after arriving in SF in my early 20s. $700 / mo, not bad for a 1 bedroom (just have to share kitchen and bathrooms with old Chinese ladies). Stayed there well after I started earning mid 6 figures. Good for the soul.
0
0
1
115
you can currently self host autopilot (harness), hotcell (sandbox) and serverless objects (durable state) on any hardware you own that runs mac or linux OS. you’ll still need to use one of the 25+ providers (open router, etc.) to access the model.
my ambitious goal is to eventually also provide the offline frontier coding model as well as the chip.
hence directionally local
my ambitious goal is to eventually also provide the offline frontier coding model as well as the chip.
hence directionally local
i’m starting a new startup in singapore and i’m calling it directionally local.
people who are serious about software should still make their own hardware.
directionallylocal.com
0
0
1
178
I believe a computing breakthrough will come out of China in the next 6–12 months, making today’s data centers look stupid.
I’m building Directionally Local for that near future.
I’m building Directionally Local for that near future.
i’m starting a new startup in singapore and i’m calling it directionally local.
people who are serious about software should still make their own hardware.
directionallylocal.com
0
0
1
191
new open source project: serverless objects (pre alpha)
“Run Cloudflare Workers and Durable Objects apps on your own hardware. Self-hosted, Wrangler-compatible, built for AI agents. Part of Directionally Local.”
github.com/sinameraji/ser…
“Run Cloudflare Workers and Durable Objects apps on your own hardware. Self-hosted, Wrangler-compatible, built for AI agents. Part of Directionally Local.”
github.com/sinameraji/ser…
0
0
0
60
Sina Meraji retweeted
steve jobs launched the iphone quoting alan kay.
sina took it literally.
sina took it literally.
i’m starting a new startup in singapore and i’m calling it directionally local.
people who are serious about software should still make their own hardware.
directionallylocal.com
1
1
25
1,777
obliterated
Now live: abliterated-model-large-flash (our version of DeepSeek-V4.1-Flash-1017). It is a cheaper alternative to our GLM-5.3-based model for product owners and defenders who need access to the most cyber-capable models.
- 1M context
- Vision
- Zero input/output retention or training
We believe paying customers should be in control, and over the past few months, leading experts and major brands have adopted our models.
abliteration.ai
0
0
1
70
Sina Meraji retweeted
sina took a bet on me early in my career and i'll forever be grateful for that.
now he's building directionally local, open-source infra to run coding agents on your own hardware. everyone go check it out!
go win @sinasanm ❤️
now he's building directionally local, open-source infra to run coding agents on your own hardware. everyone go check it out!
go win @sinasanm ❤️
i’m starting a new startup in singapore and i’m calling it directionally local.
people who are serious about software should still make their own hardware.
directionallylocal.com
1
1
20
1,067
the first time i saw this tweet i didn’t read it fully and was like “damn these guys got hacked too”. if it wasn’t for some other guy’s quote tweet i would’ve remembered jev as insecure
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/81003…
0
0
0
94
Sina Meraji retweeted
> woke up to take a pee
> critical ruby vulnerability announced
> patching at 2am
this is life now.
> critical ruby vulnerability announced
> patching at 2am
this is life now.
40
28
691
18,670
