OpenClaw-style agents face workflow reality check

The most honest read is that these systems are an orchestration layer, not a script killer. They shine when work is messy and stateful, and they disappoint when you expect deterministic behavior.

• –Setup pain is mostly prompt/tool design, not infra, and Product Hunt reviews echo the same complaints about token bloat and tool sprawl.
• –Narrow tasks work; open-ended jobs loop, lose context, and burn tokens fast.
• –The clearest wins are research/content pipelines, cron/admin automation, and internal assistants with persistent memory.
• –Coding-heavy workflows are usually better served by simpler CLIs; several commenters say agent runtimes feel like a downgrade there.
• –Security and state recovery are still the hard production problems.