Working with Sub-Agents — Obvious Help Center

Beta: Sub-agent threads are available to a limited number of workspaces. Email help@obvious.ai to request access.

This guide covers what happens when Obvious spawns sub-agent threads to handle your work — what you see, how to monitor and steer them, and what to do when something needs your attention.

What mode sub-agents run in

Every sub-thread Obvious spawns runs in auto-plus mode by default. Auto-plus gives sub-agents the same tools as the main thread — workbook access, web research, integrations, file access — so they can do real work, not just report back.

Auto-plus threads understand their role from context. A thread working on one focused piece acts as an individual contributor, staying in its lane. A thread coordinating multiple sub-threads acts as a team lead, delegating and synthesizing. Obvious sets this automatically based on how the thread is spawned — you don't configure it.

You can direct the agent to use a different mode for sub-agents when you have specific needs:

Research each of these five companies in a Fast thread

Run this analysis in Analyst mode

Fast threads are useful for simple, high-volume tasks where you want speed over depth. Analyst mode makes sense when a sub-task is purely quantitative.

What triggers a sub-agent

Obvious spawns sub-threads when your request breaks into independent pieces that run better in parallel. Common triggers:

Multiple distinct subjects: "Research these five competitors" — each competitor gets its own thread
Parallel dimensions: "Analyze revenue by region, product, and segment" — each dimension runs separately
Independent sections: "Write the executive summary, methodology, and findings" — each section is drafted in parallel
Context-heavy investigation: Exploring a large codebase or crawling documentation — this runs in a sub-thread to keep the main conversation focused

You don't need to prompt for sub-agents explicitly. If a task benefits from parallel execution, Obvious handles the split. If you want it to stay in a single thread, say so at the start:

Do this in one thread — no sub-agents

Monitoring sub-agents while they work

Sub-threads appear in the thread list on the left side of your workspace, nested under the thread that spawned them. Each shows a colored status dot:

Green (pulsing) — actively working
Green (solid) — completed successfully
Red — failed
Yellow — waiting for your input
Gray — idle or archived

Click any sub-thread to open its full conversation — every tool call, every artifact it created, every step it took. You don't have to wait for the main thread to report back to see what's happening.

Sending a message to a running sub-agent

You can message any sub-thread directly, even while it's running. Use this to course-correct, add a constraint, or provide context you forgot to include in the original request.

Click the sub-thread in the thread list to open it.
Type your message in the chat input and press Enter.

The sub-agent incorporates your message and adjusts its work from that point forward. You don't need to restart the thread or relay the instruction through the main agent.

When a sub-agent needs your input

A yellow dot means the sub-agent hit a decision point it can't resolve on its own. It stopped and is waiting for you.

Click the yellow-dot thread to open it.
Read the message — it explains exactly what it needs.
Type your response and press Enter.

The thread resumes immediately. If multiple threads are waiting at the same time, address them in any order — they don't block each other.

When a sub-agent fails

A red dot means the thread reported failure. The main thread is notified and describes what went wrong, but you get the full detail by clicking into the failed thread directly.

From there, you have three options:

Re-run from the parent. Ask the main agent to retry the failed piece with corrected or more specific instructions.
Address it in the thread. Open the failed thread and send a message with the additional context or correction. The agent picks up from where it stopped.
Skip it. If the failure was on a non-critical piece, tell the main agent to proceed without it.

Obvious doesn't automatically retry failed threads. It waits for your direction so you can decide the right path.

Controlling whether sub-agents run

The default behavior is automatic — Obvious decides when parallel execution makes sense. You can change this explicitly.

To prevent sub-agents on a specific request, tell the agent before it starts planning:

Stay in this thread — no sub-agents

To request sub-agents on something Obvious might handle in one thread:

Spawn a separate thread for each company in this list

Use parallel sub-threads for each section of this report

The agent follows your preference and confirms the approach before starting.

If a sub-thread appears stuck, behaves unexpectedly, or a status indicator doesn't match what you're seeing, reach out to help@obvious.ai.