← sources

Is anyone else finding agent harnesses less efficient than just using one strong agent + an orchestrator?

source: reddit (reddit:AI_Agents)published 2026-09-14 08:56ingested 2026-09-15 02:36

retained text

I’ve been experimenting with agent harness setups lately, and I’m starting to question whether the additional complexity is actually worth it.

In theory, having multiple specialized agents sounds great: one handles planning, another implementation, another review, another testing, etc.

But in practice, I keep running into:

* Context getting lost between agents
* Agents duplicating work
* More time spent coordinating than actually doing
* Different agents making conflicting decisions
* Constant handoffs adding latency
* Debugging the harness becoming a task of its own

I’m increasingly finding that **one capable agent with good context, tools, and a strong orchestrator/planning layer** is faster and more predictable.

Instead of:

Orchestrator → Agent A → Agent B → Agent C → Reviewer → back to Agent A

I’d rather have:

**Orchestrator → One capable agent → tools/subtasks as needed**

The agent keeps the full context and can still parallelize specific tasks when there’s a genuine reason to.

Maybe I’m using harnesses wrong, but it feels like we’re sometimes recreating organizational bureaucracy inside AI systems.

For people actually using multi-agent setups for production work: **where are you seeing a real advantage over one strong agent with an orchestrator?**

document fetched 2026-09-15 00:20 · canonical https://reddit.com/r/AI_Agents/comments/1wfy37a/is_anyone_else_finding_agent_harnesses_less