A context engineer is building a Databricks research agent that produces analyst reports where every claim must cite a verifiable source: a reviewer must be able to trace each citation back to its exact original source text. At each step, the agent calls a retrieval tool that returns several short source snippets, writes a reasoning trace explaining how those snippets support its conclusions, and appends both the snippets and the reasoning to a running context reused by later steps. As the run grows, the accumulated context pushes per-step token cost above the team's budget, so the context must be compacted to reduce token spend.
Which compaction strategy meets both the citation requirement and the need to reduce tokens?

• A. Snippet summarization: replace each retrieved snippet with a short paraphrase of what it says to save tokens, while keeping every reasoning trace in full.
• B. Conservative compaction: keep the retrieved snippets verbatim but compress the reasoning traces and drop redundant text, accepting a smaller token saving in order to preserve the source text citations depend on.
• C. Recency eviction: drop the earliest steps' snippets and reasoning as new steps arrive, keeping only the most recent context that fits the token budget.
• D. Aggressive compaction: replace each step's snippets and reasoning with a brief model-written summary before passing context to later steps, since summaries reclaim the most tokens.

A context engineer is building a customer support agent on Databricks. The agent's session-initialization routine must populate a context object before turn 1 begins; the context object is then passed to every turn handler. The agent needs the user's plan tier, region, and a few entitlement flags from the Lakebase customer_profile table in that context object. The values must be present before the first turn handler executes and may be referenced on any subsequent turn. The values do not change during a session, but could change across different sessions. The team wants to minimize per-turn Lakebase round-trips and the latency they add to turn responses.
Which retrieval approach should the engineer implement for this agent architecture?

• A. Pre-load the user's plan tier, region, and entitlement flags from Lakebase into the agent's per-session context object at session start, and reuse the loaded values for the rest of the session.
• B. Query Lakebase for the plan tier, region, and entitlement flags once and store them in a shared cross- session cache, then have each session read its values from that cache at session start.
• C. Query Lakebase for the plan tier, region, and entitlement flags during each turn that references them, so the agent pulls the current values from the table on those turns.
• D. Pre-load the user's plan tier, region, and entitlement flags from Lakebase at session start, and re-fetch them on a fixed five-minute cadence to keep the agent's copy in sync with Lakebase.

A context engineer is debugging a Databricks customer support agent that manages complex billing escalations.
The system prompt instructs the agent to:
Strictly apply the company's official refund policy.
Treat entries from the "Billing-Policies" tool as authoritative.
Treat FAQ snippets and prior chat history as non-binding guidance only.
In one multi-step trace, the engineer observes the following:
Step 1: The agent calls the "Billing-Policies" tool and retrieves the official refund policy, which clearly states that chargebacks filed more than 90 days after purchase are not eligible for refunds.
Step 2: The agent calls a "Customer-FAQ-Search" tool that returns a community-authored FAQ thread. A top- voted but outdated and unofficial answer incorrectly claims that chargebacks are fully refundable regardless of when they are filed, and the agent summarizes this in the running conversation as an established fact.
Step 3: In a later turn, the customer describes a chargeback dispute from 18 months ago. The agent reasons from its own earlier summary, approves a full refund, and makes no reference to the official policy section excluding cases older than 90 days, even though that policy text remains in the context.
Step 4: In subsequent similar cases, the agent repeatedly approves refunds for old chargebacks, citing its prior summary as justification each time.
Which context failure mode describes what is occurring in this trace?

• A. Context distraction, because the context has grown so long with policy text, FAQs, and chat history that the model loses focus on the specific 90-day rule and drifts toward irrelevant retrieved content.
• B. Context confusion, because the agent misidentifies which retrieved policy section governs chargeback disputes, applying refund rules intended for a different transaction category to the current case.
• C. Context poisoning, because a mistaken statement about refund eligibility enters the context as ground truth and is repeatedly reused in downstream steps, overriding the correct policy that is also present.
• D. Context clash, because the agent is simultaneously presented with two contradictory authoritative sources and produces an inconsistent response that partially applies each policy.

A context engineer is building a compound AI system for a hospital network with Agent Bricks Custom Agents. The system answers clinician questions about a specific patient's medication history. Patient records are stored in a Delta table that is continuously updated as new prescriptions, refills, and discontinuations are recorded, and clinicians must see the most current state of a patient's record when they submit a query.
HIPAA's minimum necessary rule requires the system to retrieve only the requesting patient's records instead of searching across the broader patient population. End-to-end latency per query must stay under two seconds.
Which retrieval strategy should the context engineer use?

• A. Use a Databricks AI Search index configured in delta-sync mode so the index automatically updates from the Delta table, and perform semantic retrieval at inference.
• B. Use just-in-time agentic retrieval by equipping the agent with a tool that executes a parameterized Delta table query scoped to the specific patient at inference time.
• C. Use a hybrid approach: precompute embeddings for all patient records nightly and supplement with just- in-time retrieval only for records modified during the day.
• D. Use pre-inference embedding-based retrieval by indexing all patient records into a Databricks AI Search index at pipeline build time and querying it semantically at inference.