Page: 'XLOOKUP Practice — Session 1'\n\nWorked through 6 practice problems from the sales table.\n\nProblem 1 (single condition lookup): got it right first try, no issues.\n\nProblem 2 (lookup with IFERROR wrapper): got it, but forgot the syntax order at first — put IFERROR outside instead of wrapping just the XLOOKUP part. Fixed it once I saw the error.\n\nProblem 3 (multi-condition, region + product): messed this up twice. Tried using two separate XLOOKUPs and couldn't combine them. Not sure how to actually do an AND-style lookup with XLOOKUP.\n\nProblem 4 (same as 3 but with dates added as a third condition): skipped this one, too confused after problem 3.\n\nProblem 5 (lookup returning multiple columns at once): got it, easier than expected once I saw you could return an array.\n\nProblem 6 (approximate match / nearest value): got the right answer but honestly just guessed the match_mode number, don't actually understand what -1 vs 1 means here.\n\nOverall feeling: comfortable with basic and multi-column lookups, still lost on combining multiple conditions in one XLOOKUP, and fuzzy on match_mode options.

File: 'excel-practice-log.csv'\nRows: 25 · Columns: week, subtopic, score_pct, attempts, avg_time_sec\n\nFull data:\nWeek 1, Basic Formulas, 65%, 8, 95\nWeek 1, Lookups, 40%, 6, 140\nWeek 1, Pivot Tables, 30%, 4, 180\nWeek 1, Conditional Formatting, 55%, 5, 110\nWeek 1, Array Formulas, 20%, 3, 200\nWeek 2, Basic Formulas, 78%, 8, 80\nWeek 2, Lookups, 55%, 7, 125\nWeek 2, Pivot Tables, 45%, 5, 160\nWeek 2, Conditional Formatting, 60%, 5, 105\nWeek 2, Array Formulas, 25%, 3, 195\nWeek 3, Basic Formulas, 90%, 9, 65\nWeek 3, Lookups, 60%, 8, 118\nWeek 3, Pivot Tables, 62%, 6, 140\nWeek 3, Conditional Formatting, 70%, 6, 95\nWeek 3, Array Formulas, 30%, 4, 190\nWeek 4, Basic Formulas, 92%, 9, 60\nWeek 4, Lookups, 58%, 8, 122\nWeek 4, Pivot Tables, 75%, 7, 120\nWeek 4, Conditional Formatting, 85%, 7, 80\nWeek 4, Array Formulas, 35%, 4, 185\nWeek 5, Basic Formulas, 95%, 10, 55\nWeek 5, Lookups, 52%, 8, 130\nWeek 5, Pivot Tables, 88%, 8, 95\nWeek 5, Conditional Formatting, 90%, 8, 70\nWeek 5, Array Formulas, 40%, 5, 178\n\nAuto-detected: Lookups peaked at 60% in Week 3 and has declined every week since (60% → 58% → 52%), the only subtopic moving backward while every other subtopic has risen steadily. Average time-to-answer for Lookups has also increased since Week 3 (118s → 130s), suggesting growing hesitation rather than a fluke.

Summary  
This chapter covers building a multi-tool conversational agent with explicit tool-routing instructions and synthesis rules, allowing it to choose the appropriate tool for each part of a request and merge their outputs into concise, accurate responses.

General domain of usage  
Educational study recap and performance analysis


Your first agent had one tool and one job. This one has two — notion_page for your notes, spreadsheet_analyze for your scores — and now it has to figure out which tool answers which part of the question.



## The Trap of More Tools

Without clear routing, models fail in predictable ways:





Calls both tools every time regardless of relevance

Uses score data when the notes already had the answer

Asks you for context instead of just checking

## How to Fix It

One clear line per tool:

> <b>Use notion_page for what was actually practiced and what felt confusing. Use spreadsheet_analyze for how scores are trending over time.`



## What You're Building

A weekly recap agent: reads your latest practice notes, checks your score trends, and tells you if they're actually pointing at the same problem, but only when they really do.

Full system prompt:

> <b>You are a weekly study recap coach. Tool routing: use notion_page to read the learner's practice notes for what was practiced and what felt confusing. Use spreadsheet_analyze to check how scores are trending across weeks. You ALWAYS call both tools before answering. Synthesis rule: don't paste tool output verbatim — condense it into your own short bullets. Output format, exactly these two sections: This Week's Notes (2 to 3 bullets on what was practiced and what felt shaky), Score Trend (the one subtopic moving the wrong direction, with actual numbers). If the notes and the score trend point at the same subtopic, say so directly. If they don't clearly connect, don't force it — report them separately instead. Never invent a connection that isn't actually supported by both sources.`



> ## Task
> 1. Click "Create Agent".
> 2. Agent name 
>    ><b>Weekly Study Recap
> 3. System prompt → paste the block above
> 4. Choose tools → check Notion Page and Spreadsheet Analyze
> 5. Save, then activate the agent for this chapter
> 6. Send: 
>    ><b>Give me this week's study recap — what I practiced, how my scores are trending, and anything that connects the two.`
> 7. Check your result: the notes flag multi-condition XLOOKUP as a guess, and the scores show Lookups declining three weeks straight — the only subtopic moving backward. If your agent treats these as two unrelated facts instead of pointing out they're the same problem, tighten the synthesis rule. If it invents a connection to something unrelated, that's also wrong.

Get hands-on experience with building a personal AI learning system using Claude, from mapping your skill gaps and writing structured prompts to building agents that read your notes and scores and recommend what to study next.

Discover how to set up a personal learning workspace, define a specific skill map for what you want to learn, and run a quick diagnostic to see exactly where you're starting from.

Explore the essentials of writing structured prompts with context, instruction, format, and constraint, then practice the same skill through quizzes, explain-it-back exercises, and real-world scenarios.

Get hands-on experience with Claude tools like notion page and spreadsheet analyze, then build single- and multi-tool agents that read your notes and scores to recommend what to study next.