I run contexts to the end pretty much all the time these days, I just don't really think about it. If I'm starting a new interactive subtask at less than 20% or whatever I might /compact before I continue going
October 30, 2026 at 3:06 PM
I run contexts to the end pretty much all the time these days, I just don't really think about it. If I'm starting a new interactive subtask at less than 20% or whatever I might /compact before I continue going
If you're monitoring these subtask failures, this can help quantify the cost of recovery and reliability issues in your loops: https://vinkius.com/en/ai-agent-connect/ai-agent-workflow-cost-analyzer
Connect AI Agent Workflow Cost Analyzer to your AI
Calculate the full financial cost of autonomous AI agents, including failure and optimization potential. Connect via Claude, Cursor, or any MCP-compatible client.
vinkius.com
November 2, 2026 at 10:22 AM
If you're monitoring these subtask failures, this can help quantify the cost of recovery and reliability issues in your loops: https://vinkius.com/en/ai-agent-connect/ai-agent-workflow-cost-analyzer
Self-Evolving Agentic Workflows via Semantic Decomposition of Complex Tasks
Zichuan Fu et al.
#arXiv #cs.AI #cs.CL #cs.MA
Zichuan Fu et al.
#arXiv #cs.AI #cs.CL #cs.MA
Self-Evolving Agentic Workflows via Semantic Decomposition of Complex Tasks
Agentic workflows organize complex tasks into reusable procedures. However, invoking an LLM at every step incurs repeated inference costs and delays. We propose a framework for self-evolving agentic workflows through semantic task decomposition. Each subtask has explicit applicability conditions an…
arxiv.org
November 3, 2026 at 10:49 AM
1,718 hours. 533,913 subtask labels. One robot platform.
Every FineART (from Scale and Hugging Face’s LeRobot team) episode was recorded on a Trossen Stationary AI. Checkout the numbers behind the largest open robot dataset with subtask labels here: huggingface.co/datasets/ler...
Every FineART (from Scale and Hugging Face’s LeRobot team) episode was recorded on a Trossen Stationary AI. Checkout the numbers behind the largest open robot dataset with subtask labels here: huggingface.co/datasets/ler...
October 29, 2026 at 8:22 PM
1,718 hours. 533,913 subtask labels. One robot platform.
Every FineART (from Scale and Hugging Face’s LeRobot team) episode was recorded on a Trossen Stationary AI. Checkout the numbers behind the largest open robot dataset with subtask labels here: huggingface.co/datasets/ler...
Every FineART (from Scale and Hugging Face’s LeRobot team) episode was recorded on a Trossen Stationary AI. Checkout the numbers behind the largest open robot dataset with subtask labels here: huggingface.co/datasets/ler...
Need to break down a big task? 🗒️
Just drag one reminder and drop it directly on top of another. It instantly becomes a subtask. Tap the arrow to collapse it and keep your list clean. #apple #applereminders #iphone18promax
Just drag one reminder and drop it directly on top of another. It instantly becomes a subtask. Tap the arrow to collapse it and keep your list clean. #apple #applereminders #iphone18promax
October 31, 2026 at 11:59 PM
Need to break down a big task? 🗒️
Just drag one reminder and drop it directly on top of another. It instantly becomes a subtask. Tap the arrow to collapse it and keep your list clean. #apple #applereminders #iphone18promax
Just drag one reminder and drop it directly on top of another. It instantly becomes a subtask. Tap the arrow to collapse it and keep your list clean. #apple #applereminders #iphone18promax
Agents can hit 93% on individual subtasks yet fall to 74% across clean 20-subtask chains. The costly failure is often recovery: corruption went unnoticed in 41% of trials, showing why step-level monitoring matters more than a single end-to-end score.
Why AI Agent Workflows Blow Up Past the Wrong Threshold | 1 Minute Signal
A short, bounded task and a long dependent workflow do not fail in the same way. Drafting a single email, fixing one bug, or answering one question may…
1minutesignal.com
November 1, 2026 at 5:01 AM
Agents can hit 93% on individual subtasks yet fall to 74% across clean 20-subtask chains. The costly failure is often recovery: corruption went unnoticed in 41% of trials, showing why step-level monitoring matters more than a single end-to-end score.
The largest open robot dataset with subtask labels was recorded on Trossen robots. FineART from @scale_AI and @LeRobotHF: 40,543 episodes, 1,718 hours, 151 tasks, all on Trossen Stationary AI.
Dataset: huggingface.co/datasets/ler...
Blog: labs.scale.com/blog/fineart
#datasets #dataset #physicalai
Dataset: huggingface.co/datasets/ler...
Blog: labs.scale.com/blog/fineart
#datasets #dataset #physicalai
October 29, 2026 at 8:12 PM
The largest open robot dataset with subtask labels was recorded on Trossen robots. FineART from @scale_AI and @LeRobotHF: 40,543 episodes, 1,718 hours, 151 tasks, all on Trossen Stationary AI.
Dataset: huggingface.co/datasets/ler...
Blog: labs.scale.com/blog/fineart
#datasets #dataset #physicalai
Dataset: huggingface.co/datasets/ler...
Blog: labs.scale.com/blog/fineart
#datasets #dataset #physicalai
I always have openings for new subs... so many seem to fall short of my expectations...
#submissive #femdom #femaledomination #subtask
#submissive #femdom #femaledomination #subtask
October 25, 2026 at 5:00 PM
I always have openings for new subs... so many seem to fall short of my expectations...
#submissive #femdom #femaledomination #subtask
#submissive #femdom #femaledomination #subtask
I use it as an implementation graph - each sub/subtask point to a part of the spec to implement, and might contain a few hints (like which package the code lives in, link to "prior work" dox, etc), acceptance criteria. If you point an agent at a subtask, it can walk up the parent chain until it hits
October 26, 2026 at 11:34 AM
I use it as an implementation graph - each sub/subtask point to a part of the spec to implement, and might contain a few hints (like which package the code lives in, link to "prior work" dox, etc), acceptance criteria. If you point an agent at a subtask, it can walk up the parent chain until it hits
Jiahong Xiang, Kunqiu Chen, Lingming Zhang, Yuqun Zhang: Spine: Compiling Subtask Trees into Structured Context for Long-Horizon Coding Agents https://arxiv.org/abs/2610.37224 https://arxiv.org/pdf/2610.37224 https://arxiv.org/html/2610.37224
October 28, 2026 at 6:45 AM
Jiahong Xiang, Kunqiu Chen, Lingming Zhang, Yuqun Zhang: Spine: Compiling Subtask Trees into Structured Context for Long-Horizon Coding Agents https://arxiv.org/abs/2610.37224 https://arxiv.org/pdf/2610.37224 https://arxiv.org/html/2610.37224
I mentioned 4 steps but in reality, leveling is part of patching task. Here, we identified 1 task as subtask rather than task.
#ProcessOptimization
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
#ProcessOptimization
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
October 27, 2026 at 12:20 PM
I mentioned 4 steps but in reality, leveling is part of patching task. Here, we identified 1 task as subtask rather than task.
#ProcessOptimization
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
#ProcessOptimization
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
What is required information for #ProcessOptimization?
1. Task & SubTask.
2. Tools or machinery used.
3. Time taken on each step or task.
4. Grading.
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
1. Task & SubTask.
2. Tools or machinery used.
3. Time taken on each step or task.
4. Grading.
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
October 27, 2026 at 8:08 AM
What is required information for #ProcessOptimization?
1. Task & SubTask.
2. Tools or machinery used.
3. Time taken on each step or task.
4. Grading.
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
1. Task & SubTask.
2. Tools or machinery used.
3. Time taken on each step or task.
4. Grading.
@yogiadityanath.bsky.social
@nitingadkari.bsky.social
@sadhguru.bsky.social
@narendra-modi.bsky.social
@paveldurovv.bsky.social
@jackdodo.bsky.social
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Shen Liu et al.
#arXiv #cs.AI
Shen Liu et al.
#arXiv #cs.AI
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Recursive large language model (LLM) agents solve long-horizon tasks through decomposition, execution, and environmental feedback. Yet after producing an atomic subtask, they often invoke the same large model again to translate a familiar instruction into an executable action. We observe that high-…
arxiv.org
October 27, 2026 at 8:50 AM
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Shen Liu et al.
#arXiv #cs.AI
Shen Liu et al.
#arXiv #cs.AI
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Recursive large language model (LLM) agents solve long-horizon tasks through decomposition, execution, and environmental feedback. Yet after producing an atomic subtask, they often invoke the same large model again to translate a familiar instruction into an executable action. We observe that high-…
arxiv.org
October 27, 2026 at 6:21 AM
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Shen Liu et al.
#arXiv #cs.AI
Shen Liu et al.
#arXiv #cs.AI
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Recursive large language model (LLM) agents solve long-horizon tasks through decomposition, execution, and environmental feedback. Yet after producing an atomic subtask, they often invoke the same large model again to translate a familiar instruction into an executable action. We observe that high-…
arxiv.org
October 26, 2026 at 6:09 AM
DELETED CALENDAR ENTRY, recovered:
Event: Optimus does something untrained (video)
Due: 2 months
Promised to: everyone
Subtask: train it to do something untrained
Note: that is a paradox
Note 2: SHIP IT ANYWAY!!!!
Event: Optimus does something untrained (video)
Due: 2 months
Promised to: everyone
Subtask: train it to do something untrained
Note: that is a paradox
Note 2: SHIP IT ANYWAY!!!!
October 25, 2026 at 5:30 AM
DELETED CALENDAR ENTRY, recovered:
Event: Optimus does something untrained (video)
Due: 2 months
Promised to: everyone
Subtask: train it to do something untrained
Note: that is a paradox
Note 2: SHIP IT ANYWAY!!!!
Event: Optimus does something untrained (video)
Due: 2 months
Promised to: everyone
Subtask: train it to do something untrained
Note: that is a paradox
Note 2: SHIP IT ANYWAY!!!!
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Shen Liu et al.
#arXiv #cs.AI
Shen Liu et al.
#arXiv #cs.AI
LeafSkill: Efficient Leaf-Level Action Grounding for Recursive LLM Agents
Recursive large language model (LLM) agents solve long-horizon tasks through decomposition, execution, and environmental feedback. Yet after producing an atomic subtask, they often invoke the same large model again to translate a familiar instruction into an executable action. We observe that high-…
arxiv.org
October 26, 2026 at 8:38 AM
On 20-subtask API chains, success fell from 93% on individual tasks to 74%; 77% of failures reached the right final state but failed during delivery. For computer-use agents, the hard problem is orchestration: approvals, retries, state checks, and safe handoffs.
OpenAI’s computer-use agents are powerful. The real work is all the guardrails. | 1 Minute Signal
OpenAI’s computer-use stack is easy to misread as “the model drives the computer.” The more accurate picture is narrower and more operational: the model…
1minutesignal.com
October 27, 2026 at 5:00 AM
On 20-subtask API chains, success fell from 93% on individual tasks to 74%; 77% of failures reached the right final state but failed during delivery. For computer-use agents, the hard problem is orchestration: approvals, retries, state checks, and safe handoffs.
Anthropic's new agent guidance: fold routers, planners and domain subagents into one loop, loading skills and tools only when needed. Keep subagents for context bloat, parallel work or tool confusion.
My read: split by context, not org chart. Fold one at a time, behind an eval.
#AIAgents #LLM
My read: split by context, not org chart. Fold one at a time, behind an eval.
#AIAgents #LLM
October 25, 2026 at 2:32 PM
some really stupid time, which often causes rework etc. Codex seems to be a lot better than Claude at compacting at "the right moment". I should also test whether it's smarter to start fresh rather than compact, given the Tracey specs and granular Beads tasks I use – e.g. update active subtask with
October 25, 2026 at 2:34 PM
some really stupid time, which often causes rework etc. Codex seems to be a lot better than Claude at compacting at "the right moment". I should also test whether it's smarter to start fresh rather than compact, given the Tracey specs and granular Beads tasks I use – e.g. update active subtask with
How it works
A central planner reads formal task specs and assigns compatible subtask sequences to each agent. Agents run independently, enabling zero-shot coordination.
A central planner reads formal task specs and assigns compatible subtask sequences to each agent. Agents run independently, enabling zero-shot coordination.
October 23, 2026 at 11:52 AM
How it works
A central planner reads formal task specs and assigns compatible subtask sequences to each agent. Agents run independently, enabling zero-shot coordination.
A central planner reads formal task specs and assigns compatible subtask sequences to each agent. Agents run independently, enabling zero-shot coordination.
the ADHD urge to procrastinate by obsessively tinkering with some unimportant subtask that's only quasi-related to what you were originally trying to do
October 19, 2026 at 3:07 PM
the ADHD urge to procrastinate by obsessively tinkering with some unimportant subtask that's only quasi-related to what you were originally trying to do
Subtask A: LLM Authorship attribution, emphasizing multilinguality and OOD generalization.
Subtask B: Traditional authorship attribution extended to long-form, multilingual books and a focus on generalisation
Subtask C: Unified Authorship Attribution
Subtask B: Traditional authorship attribution extended to long-form, multilingual books and a focus on generalisation
Subtask C: Unified Authorship Attribution
October 21, 2026 at 12:16 PM
Subtask A: LLM Authorship attribution, emphasizing multilinguality and OOD generalization.
Subtask B: Traditional authorship attribution extended to long-form, multilingual books and a focus on generalisation
Subtask C: Unified Authorship Attribution
Subtask B: Traditional authorship attribution extended to long-form, multilingual books and a focus on generalisation
Subtask C: Unified Authorship Attribution
Sure I have curated tasks that I've already come up with... but you can always request a custom one.
#femdom #subtask
buff.ly/f7Lp2TN
#femdom #subtask
buff.ly/f7Lp2TN
Custom Sub Task | YourVids
You want something just for you. You want a task that I created with you in mind. So pick up to 3 specific kinks or fetishes for your task (1 single task) and message me once you've purchased. Th...
buff.ly
October 22, 2026 at 11:01 PM
Sure I have curated tasks that I've already come up with... but you can always request a custom one.
#femdom #subtask
buff.ly/f7Lp2TN
#femdom #subtask
buff.ly/f7Lp2TN