Dots: Always-on agents
TL;DR Highlight
OpenAI's Dots launch drew mixed HN views on cost, trust, lock-in, and real use.
Who Should Read
Backend or platform developers evaluating always-on agents
Core Mechanics
- No article body was provided
- Only the title "Dots: Always-on agents" and the URL are available.
- Features, pricing, and numbers cannot be confirmed from the source.
Community Reactions
- Splitting agents by domain worked well, one user said
- A heavy Grok Bot user said separate work and personal bots keep memory from mixing.
- Combined with cloud agents, it solved running multiple git worktree streams at once.
- But tasks took about 1 hour versus about 10 minutes locally, and inference spend rose 2-3x.
- No need for overnight agents, one view
- One developer said their throughput is limited by their own approval speed.
- They said webhooks can already deliver pre-researched bug reports.
- Concerns about access and trust
- One commenter said coding agents still sometimes go the wrong way or misreport what they did.
- Giving unsupervised agents person-level access was called a footgun.
- One user claimed the cloud environment is MITM, seeing an OpenAI-issued certificate for gmail.com.
- Lock-in and pricing concerns
- One view was that integrations and work history make agents harder to switch than models.
How to Apply
- Start agents with read-only permissions
- List the connected-app permissions of the agent you use today.
- Turn off write and delete access and run read-only first.
- Separate work and personal agent contexts
- Create separate agents and memory per domain.
- Block them from reading each other's memory.
- Measure cost and time before adopting
- Run the same task locally and via an async agent.
- Record time and inference cost for each and compare.
Terminology
Related Papers
Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy
Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.
Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents
Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.
Harness Engineering 101: The LLM is the brain, the harness is the body
An agent is a loop managing a JSON array; 16 chapters build one from scratch
Harness2 by Google Research – Recursive Agent Harnessing
Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.
Ousted Savannah police officer gave ICE agent husband access to Flock system
A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired
Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling
Running Claude Code agents in tmux works well, but limits appear around four agents.