A deterministic workflow has its steps fixed in code, so every run follows the same path. An agentic workflow lets the model pick the next step at runtime, so the path changes with the input.
Deterministic workflow
Agentic workflow
Picks the next step
Your code
The model
Path across runs
Same
Varies
Testing
Assert on the output
Judge the whole trajectory
Cost and latency
Predictable
Open-ended
Best for
Known, stable steps
Branching, open-ended tasks
The trade-off is control versus flexibility. Fixed steps buy repeatable runs, cheap debugging, and a predictable bill. Agency handles cases you never listed.
Most production systems mix both: a fixed pipeline with one agentic step where the branching is unknown. Start deterministic, and add agency only where fixed logic keeps failing.
Rewriting in plainer words…
This answer doesn't lend itself to a diagram - it reads best . No credits were charged.
If an LLM routes each request to one of three fixed handlers, is that agentic? Why or why not?It sits near the fixed end. The model picks a branch once, but code defines every branch and the end. Agents keep choosing steps until they stop.
In a deterministic workflow my code fixes the steps, so every run follows the same path.
In an agentic workflow the model chooses the next step at runtime, so the path changes with the input.
With fixed steps I can assert on the output, but with an agent I have to judge the whole trajectory, and cost and latency become open-ended.
Fixed steps give you repeatable runs, cheap debugging and a predictable bill, while agency covers the cases you never listed, so the trade-off is control versus flexibility.
Most production systems mix the two, with a fixed pipeline and one agentic step where the branching isn't known up front.
My default is to start deterministic and add agency only where the fixed logic keeps failing.
Weak answers to avoid
Treats agentic as strictly betterAgency costs you repeatable runs, easy debugging and a predictable bill. Say you start deterministic and add agency only where fixed logic keeps failing.
Calls any pipeline with an LLM call agenticCalling a model inside a step doesn't make a workflow agentic. What matters is who picks the next step, your code or the model.
Frames it as either-orMost production systems mix both. Describe a fixed pipeline with one agentic step where the branching can't be known in advance.
Skips testing and costThe difference shows up in operations. Contrast asserting on outputs with judging trajectories, and predictable cost and latency with open-ended ones.
Interview lens
Likely follow-ups, what you can say, and the weak answers to avoid.
