Agentic Systems and the Transformation of Work: Pre-Course Quiz

🤖 Agentic Systems and the Transformation of Work

Pre-Course Discovery Quiz

Your quiz results will be saved anonymously for course preparation purposes.

Question 1 of 15

Question 1: What fundamentally distinguishes an agentic AI system from a traditional large language model (LLM) that simply responds to prompts?

A Agentic systems use larger neural networks with more parameters, enabling them to understand complex instructions and generate more sophisticated outputs

B The system autonomously decomposes high-level goals into executable sub-tasks, selects appropriate tools, and adjusts its strategy based on intermediate results

C Agentic AI requires specialized hardware infrastructure including GPUs and TPUs that enable faster processing speeds compared to standard models

D The system is trained exclusively on proprietary enterprise data rather than public internet data, allowing it to make domain-specific decisions

Question 2: GitHub's internal research showed that developers using AI coding assistants completed tasks 55% faster, yet code review times increased. What does this pattern most likely indicate?

A The value proposition shifted from code generation speed to architectural review, where engineers validate AI outputs against system design principles and quality standards

B Developers deliberately slow down review processes to justify their continued employment and demonstrate value that cannot be automated by AI systems

C AI-generated code contains more subtle bugs and edge cases requiring deeper scrutiny, offsetting productivity gains with increased debugging overhead

D The measurement methodology conflates different types of tasks, failing to account for complexity differences between AI-assisted and traditional development workflows

Question 3: In the ReAct (Reasoning and Acting) architecture for agentic systems, what mechanism allows the agent to recover from errors and adapt to unexpected outcomes?

A The agent maintains a persistent memory buffer that stores all previous actions and outcomes, enabling it to avoid repeating failed approaches through exact pattern matching

B After each action, the agent observes environmental feedback, generates reasoning traces about the outcome, and uses this context to inform subsequent action selection

C The system runs multiple parallel reasoning chains simultaneously and selects the most promising path using a discriminator model trained on successful trajectories

D Human operators monitor the agent's decision-making in real-time and provide corrective feedback when the system begins pursuing suboptimal action sequences

Question 4: Klarna's AI assistant reportedly handles the workload of 700 customer service agents. When evaluating this claim's strategic implications, what analysis is most critical?

A Comparing the AI system's total cost of ownership including development, maintenance, and compute infrastructure against the fully-loaded cost of human agents

B Examining whether "equivalent workload" measures simple transaction volume or accounts for complexity distribution, customer satisfaction, and problem resolution quality

C Assessing the replicability of these results across different industries given variations in customer interaction patterns and domain-specific knowledge requirements

D Investigating whether the 700-agent figure represents actual headcount reduction or theoretical capacity, and how displaced workers were redeployed within the organization

Question 5: An insurance company implements anomaly detection to flag unusual claims for human review. The system escalates 15% of claims. What optimization strategy creates the highest risk?

A Progressively lowering the escalation threshold to reduce human workload, potentially allowing sophisticated fraud patterns to slip through automated approval

B Incorporating historical approval patterns into the training data, which may perpetuate systematic biases in claims processing across demographic groups

C Implementing parallel human review systems to validate AI decisions, creating redundant processes that negate the efficiency gains of automation

D Requiring the AI system to provide natural language explanations for each escalation decision before allowing human adjusters to review flagged claims

Question 6: A management consulting firm restructures its analyst workflow: AI conducts initial market research (15h → 2h) and data analysis (15h → 3h). What represents optimal human time reallocation?

A Reducing project timelines proportionally while maintaining the same scope, allowing the firm to increase client throughput and revenue per consultant

B Expanding analytical depth, strategic synthesis, and client-specific contextualization that differentiates premium consulting from commoditized research reports

C Implementing comprehensive quality assurance processes to verify AI-generated insights, ensuring accuracy before presenting findings to clients

D Developing AI prompt engineering capabilities and tool orchestration skills to maximize the effectiveness of human-AI collaboration systems

Question 7: After deploying an agentic customer service AI, metrics show: average handle time -65%, CSAT unchanged, first-contact resolution -12%, escalations +35%. What diagnosis is most accurate?

A The system successfully processes routine queries but lacks contextual reasoning for complex issues, creating customer frustration through repeated interactions

B Customers have learned to manipulate the system to reach human agents faster, gaming the escalation triggers to bypass automated responses

C The deployment is successful by efficiency metrics; declining resolution rates reflect acceptable tradeoffs between speed and comprehensiveness in support delivery

D Human agents now handle exclusively complex cases requiring longer resolution times, which statistically distorts the first-contact resolution metric

Question 8: When an agentic system is given access to external tools (APIs, databases, calculators), what architectural component determines which tool to invoke at each decision point?

A A separate classifier model trained specifically on tool-selection tasks evaluates the current state and outputs the optimal tool choice from the available set

B The language model generates both natural language reasoning about which tool is appropriate and structured output specifying the tool name and parameters to invoke

C A rule-based system maps input keywords and task patterns to pre-defined tool sequences, with the agent executing the matched workflow template

D The system cycles through all available tools in parallel and selects the result that maximizes a learned utility function based on output quality metrics

Question 9: A senior engineer argues their role will be obsolete when AI can generate full-stack applications from requirements. What capability represents the most defensible competitive advantage?

A Deep expertise in legacy system architectures and technical debt patterns that AI systems cannot access through training data alone

B Translating ambiguous business requirements into well-specified technical problems, evaluating AI-generated solutions against unstated constraints, and making trade-off decisions under uncertainty

C Relationship management with stakeholders and clients who require human interaction for trust-building and high-stakes technical decision validation

D Specialized knowledge of emerging programming paradigms and cutting-edge frameworks that haven't accumulated sufficient training data for AI mastery

Question 10: A legal firm deploys AI for contract analysis. Attorneys report unchanged hours per contract despite 10x AI processing speed. Before concluding failure, investigate:

A Whether attorneys now handle higher-value contracts or perform deeper risk analysis, representing quality improvement rather than productivity failure

B If billable hour incentives create perverse motivation to maintain time expenditure despite efficiency gains from AI augmentation

C Whether measurement methodology captures total time including AI review and correction, which may offset initial processing speed advantages

D The possibility that contract complexity increased coincidentally with AI deployment, confounding comparisons to baseline productivity metrics

Question 11: Chain-of-thought prompting improves agentic system performance on complex reasoning tasks. What is the primary mechanism by which this works?

A Breaking problems into intermediate steps allows the model to allocate more computational resources to each sub-problem, increasing the effective depth of reasoning

B Explicit reasoning traces provide additional context tokens that prime the model's attention mechanisms toward solution-relevant patterns from training data

C The structured format forces the model to generate outputs that can be parsed by downstream verification systems, enabling automatic error detection and correction

D Humans reviewing the reasoning chains can provide targeted feedback on specific logical errors, creating a supervised learning signal that improves future performance

Question 12: In designing an agentic financial audit system, where does irreducible human judgment create the most critical value-add versus AI capabilities?

A Executing standardized testing procedures and variance analyses across transaction populations to identify statistically significant anomalies

B Interpreting anomalies within strategic business context, assessing materiality through professional skepticism, and distinguishing fraud from legitimate complexity

C Extracting and normalizing financial data from diverse source systems while ensuring compliance with evolving regulatory reporting standards

D Documenting findings in standardized audit workpapers and generating preliminary reports that conform to professional association formatting requirements

Question 13: Company A achieves 40% productivity gains from agentic AI; Company B sees 5% with identical technology. Organizational research attributes the divergence primarily to:

A Differences in employee age demographics, with younger workers demonstrating higher comfort levels and adoption rates for AI collaboration tools

B Company A's baseline processes contained more automation-amenable tasks, while Company B's work involved higher proportions of irreducible human judgment

C How leadership framed AI deployment—Company A emphasized augmentation enabling role evolution; Company B's replacement narrative triggered defensive resistance

D Company A's substantially larger implementation budget enabled superior change management, training infrastructure, and technical support resources

Question 14: Long-running agentic systems require memory to maintain context across extended interactions. What distinguishes episodic memory from semantic memory in these architectures?

A Episodic memory stores specific interaction sequences and conversation histories, while semantic memory contains distilled knowledge and learned task patterns extracted across experiences

B Episodic memory uses vector embeddings for similarity-based retrieval, while semantic memory employs graph structures to represent hierarchical knowledge relationships

C Episodic memory persists only during a single session and resets between interactions, while semantic memory accumulates permanently across all system deployments

D Episodic memory is optimized for rapid access to recent context, while semantic memory requires computational overhead for retrieval and is queried only when explicitly needed

Question 15: Two candidates possess identical domain credentials. Candidate B additionally demonstrates AI problem decomposition, critical output evaluation, and rapid capability adaptation. Economically, what premium should B command?

A Minimal premium—AI orchestration skills represent easily trainable capabilities that organizations can develop internally through standard upskilling programs

B Substantial premium—the capacity to multiply impact through effective AI leverage will increasingly differentiate high-performers as automation handles routine expertise

C Negative premium—excessive AI focus signals insufficient depth in core domain, creating liability as responsibilities expand beyond tool-mediated work

D Context-dependent premium—relevant for technical roles requiring AI integration but irrelevant for strategic positions demanding primarily human judgment

🎉 Quiz Complete!

0/15

💡 What You'll Explore in the Course:

Agentic AI Architecture: Understanding autonomous planning, tool use, and multi-step workflow execution
Goal Alignment: Designing objectives that reflect strategic value, not just operational metrics
Workflow Redesign: Strategic frameworks for optimal human-AI task allocation
Success Metrics: Moving beyond speed to measure quality, value creation, and failure modes
Change Management: Why identical technology produces 10x different outcomes
Career Strategy: Building defensible skills combining domain expertise with AI orchestration

Incorrect Answer

Correct Answer:

Continuing to next question in: 15