← Back to AI agents
All 10 questions. After you submit, each question shows the right answer and why. Back to the quick check
Why might an AI agent use a calculator tool instead of working out a sum by itself?
An AI agent doesn't just answer a question — it can take actions, see what happened, and then decide what to do next.
Each step of a 5-step task succeeds 90% of the time, independently. What is the chance (as a decimal, 2 decimal places) that all five steps succeed?
What does the ReAct method have a language model do?
Instructions hidden inside a web page that an AI agent reads can change what the agent does.
According to the AI Index 2026, roughly what task-success accuracy had the best agent reached on OSWorld?
Which four modules make up the unified agent architecture proposed in Wang et al.'s survey of language-model agents?
By how many percentage points of absolute success rate did ReAct outperform imitation- and reinforcement-learning baselines on ALFWorld?
Give two concrete harms NIST says researchers have demonstrated through indirect prompt injection, and one design control an agent builder can use to limit them.
NIST's framework assumes every AI system operates with the same, fixed level of autonomy.
