Most 'AI agents' are if-else workflows with an LLM bolted on — here's a quick test
dumpshoot · reddit · 2026-09-25
A Reddit post argues most products marketed as AI agents are just fixed workflows with an LLM plugged in for text generation or intent classification.
Quick test: hit the system with inputs it has never seen — a missing field, a tool erroring mid-run, a user request contradicting the original instructions. If it adjusts its plan, picks a different tool, or asks a clarifying question, it's an agent. If it crashes, goes off script, or repeats the same response, it's a workflow in an agent costume.
The boundary: a true agent runs a plan-act-observe-replan loop; a workflow is a fixed sequence where the LLM never decides action order or skips steps.
Workflows aren't bad — they're more stable, easier to debug, cheaper on tokens, and the right call for support bots, lead qualification, and content pipelines where predictability beats autonomy. The problem is selling them as agents and letting customers expect edge-case handling.
Mental model: ask "what happens if the first tool call fails?" Crash or fixed error = workflow; LLM tries something else = agent.
More from coding & agent
- Theo: $200 Claude Code plan now clearly beats Codex, weeks after trailing it — ssh4net · 2026-09-25
- Containers Aren't a Real Security Boundary: Kata Containers and Firecracker Urged for Sandboxes — andreamichi · 2026-09-25
- Reddit asks: what's the real advantage of routing smaller model jev into LLM prompts — sogo00 · 2026-09-25
- Dev ditches Amp over subscription limits, tries Claude Desktop cloud sessions — iannuttall · 2026-09-25
- One prompt, 62 seconds, under 5 cents: Xiaomi MiMo V2.6 Pro builds a full habit tracker in Claude Code — socialwithaayan · 2026-09-25
- Coding Agents Beat Hand-Engineered Planners at Generalized TAMP, 56%-95% vs 47% — FBK-NLP · 2026-09-25