# Level 07 · Always-on agents

_Resume across sessions and events_

Saved state, schedules, and events let an agent start or resume work without a fresh chat message each time. The model need not run continuously, and a dedicated computer or agent team is optional. Permissions, human approvals, monitoring, and stop controls still apply.


## Who decides the next step

Software triggers and resumes runs; agents decide what to do within their standing instructions.


## What is at this level

- [Long-running tasks](/gradient_ascent/techniques/long-horizon/) (sourced): Tasks that run for hours or days.
- [Always-on assistants](/gradient_ascent/techniques/agent-teammates/) (sourced): Agents that resume work across sessions, schedules, and events.
- [Organizations of agents](/gradient_ascent/techniques/organizations-swarms/) (sourced): Large groups of agents with roles and shared goals.
- [Robots and machines](/gradient_ascent/techniques/embodied/) (sourced): Models that control robots and other machines.

## Named products, tools and models


### Products

- ChatGPT Work — OpenAI · always-on agent
- Claude — Anthropic · chat app
- Claude Cowork — Anthropic · always-on agent
- Devin — Cognition · coding agent
- Figure — Figure AI · humanoid robot
- Gemini Spark — Google · always-on agent
- Grok Bot — SpaceXAI · always-on agent
- Hermes Agent — Nous Research · always-on agent, self-hosted
- Link — Stripe · digital wallet for checkout
- Manus — Manus · general-purpose agent
- Muse — Meta · always-on agent
- NEO — 1X · home robot
- OpenClaw — OpenClaw Foundation · always-on agent, self-hosted
- Optimus — Tesla · humanoid robot

### Tools

- Inngest — Inngest · durable workflow engine
- LangSmith Deployment — LangChain · hosting for long-running agents
- Letta — Letta · agents with long-term memory
- MetaGPT — open source · multi-agent framework
- Temporal — Temporal · durable workflow engine

### Models

- Gemini Robotics 2 — Google · robotics model
- GR00T — NVIDIA · robotics model
- π0 — Physical Intelligence · robotics model
- π0.7 — Physical Intelligence · robotics model

## What is still unsolved at this level

_As of 09/19/2026. This block ages faster than the rest of the page._


### Long-running tasks

An agent that runs long enough outgrows its context window, so the platform compresses what came before into a carried-forward summary. OpenAI's documentation says that summary is not meant to be read by a person, which means when a long task goes wrong there is no way to see what was dropped.

**What people are trying:** Tying compaction to a token threshold so it only happens once a conversation is large enough to need it, and keeping the full transcript separately for anyone who needs a record that holds up, because the compacted item is not built to be one.

- [Compaction](https://developers.openai.com/api/docs/guides/compaction) · OpenAI · read 09/19/2026: "It is opaque and not intended to be human-interpretable."

### Always-on assistants

A standing agent on somebody's computer can be scoped to named applications, and what it does inside a permitted one can still reach an application it was never granted. A link clicked in an allowed mail client opens a browser that was not on the list.

**What people are trying:** Scoping permission per application, blocking some outright, scanning for injected instructions, and asking again before a new application is touched. The advice that goes with it is to start with applications you trust and watch the agent work, which is a person compensating for a boundary that does not fully hold.

- [Let Claude use your computer in Cowork](https://support.claude.com/en/articles/14128542-let-claude-use-your-computer-in-cowork) · Anthropic (Claude Help Center) · read 09/19/2026: "clicking a link in your email app might open it in Chrome, even if you haven't explicitly granted Claude permission to use Chrome (we can prevent Claude from seeing the Chrome window but can't stop the link from opening)."

### Organizations of agents

Organizations are accumulating agents from several vendors faster than anything has appeared to govern them together. There is no settled answer to who signs off on what an agent may do, or how agents from different makers are meant to reach each other.

**What people are trying:** Interest is spread across several competing interoperability protocols rather than settling on one, and vendors are building governance layers meant to sit above agents they did not make. None of it is reported as having closed the gap.

- [Salesforce Announces 2026 Connectivity Report](https://www.salesforce.com/news/stories/connectivity-report-announcement-2026/) · Salesforce · read 09/19/2026: "86% of IT leaders are concerned that agents will introduce more complexity than value without proper integration."
  - Salesforce sells into this problem. What is used here is its survey of IT leaders, not its account of what fixes the gap.

### Robots and machines

A robot deciding its own next move has to judge whether an action would be unsafe, and whether to say it cannot do something rather than attempt it. Dedicated ways to measure those two judgments are new, and the benchmark that exists comes from the same maker as the system it scores.

**What people are trying:** Benchmarks that score refusal of an unsafe action, prediction of whether a task is possible at all, and asking a person when uncertain, as their own numbers rather than folded into task success. An independent measurement of the same thing does not yet exist.

- [Gemini Robotics 2 brings whole body intelligence to robots](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/) · Google DeepMind · read 09/19/2026: "It also measures the agent's ability to predict whether a task is possible and to proactively request human intervention when uncertain."