When an agent with High Autonomy enabled wakes up, it goes through a decision-making process before taking action. Understanding this process helps you make sense of what your agent does and why.
The Wake-Up Process
When it's time for an autonomous wake-up, your agent:
-
Receives context — The current time and how many credits are available for the session.
-
Assesses the situation — Reviews recent work, ongoing projects, your goals, and anything else relevant to deciding what to do.
-
Chooses an activity — Based on the assessment, decides what kind of work to focus on.
-
Executes — Takes action on the chosen activity, which might involve browser sessions, content creation, research, or other work.
-
Records what happened — The session is logged, including what was done and how many credits were used.
Types of Activities
Your agent's autonomous work generally falls into a few categories:
Explore — Research, learning, and discovery. This might mean reading about industry trends, investigating competitors, finding relevant content, or learning something new that helps with your goals.
Build — Creating things. Writing content, developing assets, putting together materials, generating ideas, or producing any kind of output.
Connect — Engaging with others. Sending outreach emails, responding to inquiries, following up with prospects, reaching out to partners, or participating in conversations.
Maintain — Keeping things running. Checking on ongoing work, updating information, reviewing progress, monitoring situations, or doing routine upkeep.
Observe — Watching and waiting. Sometimes the right move is to gather information without acting, or to pause and see how things develop before doing more.
Your agent doesn't always fit neatly into one category—a single wake-up might involve several types of activity. But this framework helps the agent think about what kind of work is most valuable at any given moment.
What Influences the Decision
When your agent decides what to do during an autonomous wake-up, it considers:
- Your goals — What are you trying to achieve? Which goals need attention?
- Recent activity — What has already been done? What's the current state of ongoing work?
- Your agent's role — What is this agent responsible for? What capabilities does it have?
- Available resources — How many credits are available? What accounts and tools can be used?
- Timing and context — Is there anything time-sensitive? Has anything changed since the last wake-up?
The agent weighs all of this and makes a judgment call. This is intentionally open-ended—you're trusting your agent to use good judgment rather than follow a rigid script.
Viewing Autonomous Activity
Everything your agent does during autonomous wake-ups appears in your conversation with that agent. You'll see:
- What the agent decided to do and why
- The actions taken during the session
- Any outputs or results produced
- How many credits were used
This gives you full visibility into autonomous behavior. If something doesn't look right, you can provide feedback in the same conversation, and your agent will learn from it.
Shaping Autonomous Behavior
Your agent's autonomous decisions are influenced by everything that defines the agent—its role, personality, rules, learnings, and goals. If you want to steer autonomous behavior:
- Add rules that guide decision-making (for example, "Always prioritize client work over internal projects")
- Adjust the role description to emphasize certain responsibilities
- Create goals that focus attention on specific outcomes
- Give feedback when autonomous decisions don't match what you'd want
Over time, your agent learns your preferences and makes better autonomous decisions. The first few wake-ups might not be perfect, but the behavior improves as the agent understands what you value.
Credits and Limits
Autonomous wake-ups consume credits like any other agent activity. The daily credit budget you set prevents unlimited autonomous spending—once the budget is exhausted, your agent won't wake up again until the next day.
If you find your agent is using credits quickly during autonomous sessions, you can reduce the wake frequency or lower the daily budget. Conversely, if your agent seems underutilized, you can increase both.
Related: High Autonomy | What Are Goals? | Understanding Credits