Showrunner
meta-muse-wonderrun 13 Oct 2026356,041 tokens
Instructions
Instructions — what the model was asked
Hi, I need to create a modified Claude code for beginners session, like run sheet basically, run of show, that has the following modifications. So I'm going to upload a few different examples of things I've done in the past, just so you have them to work from. But basically, there's a few modifications for this one which make it unique. One is that this is only using Claude Cowork, so we should...
Hi, I need to create a modified Claude code for beginners session, like run sheet basically, run of show, that has the following modifications. So I'm going to upload a few different examples of things I've done in the past, just so you have them to work from. But basically, there's a few modifications for this one which make it unique. One is that this is only using Claude Cowork, so we should not be using Claude Code, either in the desktop app or in the terminal, because this team is non-technical. The team are all working in financial services slash accounting for a education company. So what kind of workplace education company called Brightmoor Education. So the sorry that spelled b the uh team is very busy and um doesn have a lot of like time to explore speculative use cases of ai they want this to be incredibly practical and based on their existing workflows. I'm uploading some transcripts from a call we had with them and also just some notes on what I think could work. Specifically, they're interested in... creating the dashboards and specifically they mean just making the information look visibly better like like creating a html dashboard in a nice kind of style from from existing say data that contained in Excel They also interested in doing financial analysis So more of a, you know, like, I guess, a cloud coworker would use Python to do the analysis and then maybe output the analysis in Excel or the input would be in Excel. And then the third thing was like making a PowerPoint where they update the PowerPoint template automatically using Claude Cowork so that they can do kind of weekly or monthly presentations on the data. So that's the goal, is to kind of make it very focused on that. I still want to keep the way that we introduce Claude Cowork, where we explain to it that the difference between Claude Cowork and ChatGPT is that it has access to the file system. It can run code. I think that important But maybe the ASCII image idea that I normally do where I ask it to create an ASCII image based on my location isn like that useful I think we need to make it more practical for them So maybe we use an Excel implementation instead. The assumption is... oh, so they use Snowflake, by the way, for all their data, and that is powered by NetSuite, which is our accounting software. I guess that's by Oracle. So when we create synthetic data, we'll be, for these tasks that we're going to demo, we'll be focused specifically on solving that, solving those problems. So like the shape of the tasks, the shape of the data should be like NetSuite shaped, if that makes sense. So make it very tight run down, make it very condensed and practical. No real like wider theory or like, hey, we're excited about AI things. I think these guys just want to get stuff done.
So yeah, let me know what the run should be.
some very specific asks in here
- definitely non-technical (claude cowork only + claude in excel)
- core focus on accounting/finance function
- use cases to demo: creating a dashboard, doing financial analysis, making a powerpoint
heavy focus on practical stuff rather than theory - they're just super busy and want to get stuff done, have heard from friends at other companies you can do it with claude
For accounting: put financials together, doing reconciliations, identifying trends in data sets, putting together repeatable compliance reports For planning & forecast: build financial models (make assumptions and pull & clean data from snowflake)create dashboards, and publish to executivesAutomate charts on slides
Claude Cowork for Finance — Brightmoor Education
Modified Run of Show: Non-Technical, Cowork-Only, 3.5 Hours
Audience: 20–30 Finance / Accounting / Payroll / Bursar staff, Brightmoor Education (workplace education company) + Strategic Finance (Dana's team)
Level: ~90% beginners. Dana + 2–3 others have used Claude / Claude in Excel before — do not use them as TAs (too busy, per Dana).
Hard constraints for this version: Claude Cowork only (Desktop app) + Claude in Excel. No Claude Code. No Terminal. No GitHub. No deployment/hosting theory.
Format: 3.5-hour session (4-hour room hold, 30-min buffer). Target: ≥50% hands-on. Every build uses files they already recognise.
Data story (tell it once, up top): NetSuite → Snowflake → Excel export → [Dashboard / Analysis / PowerPoint]. Today we start from the Excel export so nobody is blocked on Snowflake provisioning. The same prompts work when the Snowflake connector is live — only the first line changes ("pull from Snowflake" instead of "read this file").
Files in this pack (share via one folder link before the session):
brightmoor-data/— synthetic, NetSuite-shaped data (seeREADME-data.md)brightmoor-prompts.md— copy-paste prompt handout (participants keep this)- This run sheet (facilitator only)
What changed vs. the standard Claude Code for Beginners / Nvalo runs, and why
- Cowork replaces Code everywhere — no Terminal/GitHub/Plan Mode jargon. The mental model is "a colleague with access to this folder who can run code," not "a coding tool."
- First prompt swapped: ASCII map → a 3-minute Excel clean-up on a real NetSuite export. Same "it can act, not just answer" wow, but on their data.
- Three builds = Dana's three asks, nothing else: (1) HTML dashboard from Excel, (2) financial analysis in Excel (Python under the hood, Excel out), (3) monthly PowerPoint from a template. Vacation planner / generic tools / open-ended "build anything" are cut.
- Theory cut to ~15 minutes total. Talk → Jam → Make → Tune and the ASK Ladder are kept but taught inside the builds, in finance language, not as frameworks up front.
- Trust & checking gets its own slot (Segment 6). Dana flagged accuracy as the #1 question once output becomes a deliverable. This is where the session earns permission to be used on real work.
- No showcase theatre. Busy team, no TA volunteers. Shares are 2-minute neighbour/pair shares + one volunteer demo per build. Ends with "pick your Monday-morning task," not a roadmap to agents/MCPs.
Pre-work (send 3 days before — non-negotiable)
Setup failure in the room is the #1 killer for this audience (per Dana call). All of this must be done before Thursday:
- Claude Desktop app installed, signed in, Cowork enabled for all 20–30 attendees (confirm with Greg / IT — single biggest risk; only "a handful" have Claude today, rest are on ChatGPT/Gemini)
- Claude in Excel add-in installed and signed in on each machine
- Download the
brightmoor-data/folder to Desktop (do not rename files) - Open Cowork once, point it at that folder, send any prompt (e.g. "list the files in this folder") — confirms permissions work
- If IT can provision the Snowflake connector in time: great, Dana demos with it in Segment 8. If not: no problem, session runs entirely on the Excel exports. Do not let this block scheduling.
- Facilitator: pre-run every prompt in
brightmoor-prompts.mdon the synthetic data; keep a known-good output dashboard + deck in the folder as fallback if live builds stall
Room setup: Tables of 3–4 (mixed Accounting / Payroll / Strategic Finance), one screen, facilitator laptop mirrored. Print the prompt handout — people follow along faster on paper.
Run of Show (3.5 hours)
Timings assume a 4-hour hold: run 0:00–3:30, hold 3:30–4:00 as overflow / 1:1 help.
0:00–0:10 — Welcome: "This replaces the 8-hour job" (10 min)
Goal: Earn attention in the first two minutes. No AI theory, no "exciting times."
Facilitator script (paraphrase freely):
"Dana's team spends about 8 hours a month turning a NetSuite export into a 15-page PowerPoint. Today, in 3.5 hours, you'll do the three pieces of that job yourselves: turn the export into a dashboard you can read, run the analysis behind it in Excel, and rebuild the monthly deck from the template. Everything today starts from files you already have. By the end you'll have done each one once, with prompts you keep."
- Introductions: name, team (Accounting / Payroll / Bursar / Strategic Finance), one repetitive report you own.
- Ground rules: work in pairs; if stuck for >2 minutes, raise a hand — don't silently fall behind; never paste real student/financial data into anything today — synthetic data only (their Snowflake/NetSuite access comes later, with IT).
- State the data story once: NetSuite → Snowflake → Excel → today. Put the diagram on screen and leave it there.
0:10–0:25 — Cowork in 15 Minutes: Not ChatGPT With a New Logo (15 min)
Goal: The one concept that matters: Cowork acts on your files and runs code. Everything else is detail.
Live demo only, no slides except one comparison:
| ChatGPT / chat Claude | Claude Cowork | |
|---|---|---|
| Sees | What you paste | The folder you point it at |
| Does | Answers, suggests | Reads files, runs Python, creates/edits files |
| Output | Text you copy out | Finished Excel / HTML / PowerPoint files in your folder |
| Checking | You re-type it | You open the file it made and check the numbers |
Demo (replaces the ASCII-art opener): Open Cowork on the brightmoor-data/ folder. Run Prompt 1 from the handout live:
"Read
netsuite-gl-export-mar.csv. Tell me: how many rows, what date range, which accounts and campuses are in it, and are there any blank or duplicate rows? Don't change the file."
Then Prompt 2:
"Now clean it: save a copy as
gl-clean.xlsxwith proper column headers, dates as dates, amounts as numbers, and a new column that flags any transaction over $10,000. Tell me what you changed."
Point at the new file appearing in the folder. That is the whole pitch: it read 2,000 rows, ran code, and handed back a workbook — nobody wrote a formula.
- Cover in 60 seconds each, only because they'll hit them today: (a) point Cowork at one folder — that's its whole world; (b) it will ask before overwriting — say "make a copy" when unsure; (c) Claude in Excel = same idea inside a workbook, for when you want formulas you can inspect cell-by-cell (used in Build 2).
- Do NOT cover: Terminal, GitHub, Plan Mode, agents, MCPs, publishing/hosting. If asked about sharing dashboards externally: "Cowork makes the HTML file; sharing it via your usual internal link/SharePoint is a 5-minute IT question after today — we'll show the file, not the hosting."
0:25–0:30 — Pair Setup Check (5 min)
Everyone opens Cowork on the data folder and runs Prompt 1 themselves. Facilitator + any confident early finishers walk the room. Do not start Build 1 until every pair has a response back. This 5 minutes saves 30 later.
0:30–1:20 — Build 1: The Executive Dashboard (50 min)
The ask (from Dana, verbatim need): "Build dashboards without a technical background… make it more readable… not relying on formatting things in Excel." Audience: senior leadership. Left-hand side = campuses. Then Dana's follow-up skills: change chart types, add filters, iterate.
Data: netsuite-gl-export-mar.csv (or the clean gl-clean.xlsx from §0) — revenue & spend by campus, by month, NetSuite-shaped (Account, Campus/Subsidiary, Department, Period, Amount).
Teach the loop by doing it, naming it only at the end:
- Talk (10 min, pairs): In Cowork, Prompt 3 — describe the dashboard before building: audience (senior leadership), the 4 questions it must answer (revenue vs. budget by campus; spend by category; month trend; anything over budget), style ("clean, modern, printable on one screen"). Have Cowork reply with a short plan/sections list. Pairs edit the plan in plain English ("drop the pie chart, add a campus filter"). This is the habit that prevents bad first drafts.
- Make (15 min): Prompt 4 — "Build it as a single HTML file called
brightmoor-dashboard.htmlin this folder." While it builds, facilitator narrates what it's doing (reading the CSV, aggregating with Python, generating charts). Pairs open the file in a browser when done. - Tune (15 min): Each pair must make three specific changes using Prompt 5 patterns — this is Dana's exact gap ("convert this to a horizontal bar chart… add filters"), so drill it:
- Change one chart type (e.g. vertical → horizontal bar)
- Add one filter (campus or month dropdown)
- Fix one readability issue (sort order, £/$ formatting, title, colour)
Rule taught here: one change per prompt, say what "good" looks like ("largest campus first," "amounts in $k, no decimals").
- Check (5 min): Prompt 6 — "Pick the revenue total for [campus] in March. Show me the rows from the source file that add up to it." Pairs verify one number against the CSV. (Seeds Segment 6.)
- Share (5 min): One pair demos; facilitator names the loop on screen: Talk → Jam (the back-and-forth) → Make → Tune. "You just did the whole method. We'll reuse it twice more."
Facilitator notes: Expect: file-permission prompts (say yes for this folder), one pair pointing Cowork at the wrong folder (fix: re-point, don't restart), chart requests that are vague ("make it nicer" → coach to a specific change). Fallback: open the pre-built dashboard and Tune from there.
— 5-minute break (1:20–1:25) — stretch, keep dashboards open on screens.
1:25–2:15 — Build 2: Financial Analysis in Excel (50 min)
The ask: "How to do analysis" + accounting specifics: put financials together, reconciliations, identifying trends, repeatable compliance reports; Strategic Finance: build financial models (assumptions + pull & clean data), forecasts. Constraint from Dana: output must be checkable — so this build lives in Excel, where every number is a formula they can audit. Cowork does the heavy lifting (Python), Claude in Excel keeps it inspectable.
Data: netsuite-gl-export-feb-mar.csv (two months) + bank-rec-mar.csv (for reconciliation) + budget-fy26.xlsx (budget vs. actuals).
Three stations — pairs do all three, in order (each ~12 min + intro/check):
- Intro (5 min, facilitator): When to use which: Cowork when the job is "take these files, do 500 steps, hand me a workbook" (cleaning, reconciling, modelling). Claude in Excel when you're already in the workbook and want a formula/explanation in place ("write a formula that flags…", "explain this variance"). Show both open side by side once.
- Station A — Reconciliation (Accounting/Payroll core): Prompt 7 in Cowork: match
bank-rec-mar.csvto the GL cash account, outputreconciliation-mar.xlsxwith three tabs: Matched / In GL not bank / In bank not GL, amounts and dates shown. Pairs open it and spot-check two matched lines against source. Payroll variant on the handout (Prompt 7b: payroll register vs. GL payroll postings) for Payroll/Bursar pairs to try instead. - Station B — Trends & variance (everyone): Prompt 8: month-on-month and vs.-budget variance by account and campus, flag anything >10% or >$25k, add a one-line plain-English explanation per flag, output as a new tab with conditional formatting. Then switch to Claude in Excel: select the variance column, Prompt 9 — "Explain the three biggest variances as if briefing a non-finance manager. No jargon." Contrast the two outputs: workbook for the record, in-Excel for the explanation.
- Station C — Simple forecast model (Strategic Finance flavour, everyone attempts): Prompt 10: "Using Feb–Mar actuals and
budget-fy26.xlsxassumptions, project Apr–Jun revenue and payroll cost by campus. Put every assumption in a clearly labelled Assumptions box at the top — I will change them. Use formulas, not hard-coded values, so I can re-run it." Pairs change one assumption (e.g. enrolment +5% at one campus) and watch it flow. This is the "repeatable" lesson: the deliverable is a model they re-run monthly, not a one-off answer. - Check (3 min): Prompt 11 — "Show me the formula behind [cell]. Walk me through it in plain English." If a pair can't explain a number, they don't ship it. (Bridge to Segment 6.)
Facilitator notes: Numbers will differ slightly between pairs if they cleaned data differently — that's a teaching moment, not an error: "Which cleaning choice caused that? That's exactly what you document." Insist outputs land as new files/tabs; nothing overwrites the source export.
2:15–3:00 — Build 3: The Monthly PowerPoint, From Template (45 min)
The ask (Dana's 8-hour job): 15-page monthly deck, NetSuite/Excel data → charts → PowerPoint, same template every month. Sub-skills Dana named: "mirror this chart and do six more," "translate these tables in Excel into PowerPoint," automate charts on slides.
Data: Pairs' own outputs from Builds 1–2 + brightmoor-monthly-template.pptx (a scrubbed, Brightmoor-styled 6-slide template: title, exec summary, revenue by campus, spend by category, variance highlights, appendix table) + deck-source-mar.xlsx.
- Mirror the chart (15 min): Prompt 12 — Cowork reads the template, takes the existing campus-revenue chart style (colours, fonts, number format), and generates the same chart for the other campuses/categories into a new deck
brightmoor-mar-deck.pptx. Pairs open it and compare slide 3 vs. the new slides side by side. Teach: "match the template" is a prompt, not a hope — name the slide it should copy. - Tables → slides (15 min): Prompt 13 — take the variance table from Build 2, produce one summary slide (top 5 variances, traffic-light status, one-line commentary each) + keep the full table in the appendix. Commentary drafted by Claude, edited by the pair — facilitator rule stated out loud: "Claude drafts commentary, a human owns it."
- The monthly rerun (10 min): The payoff. Prompt 14 — pairs save the whole sequence as a reusable instruction (a
monthly-deck-instructions.mdin their folder: source files → cleaning → charts → slides → checks). Demo the rerun live: swapdeck-source-mar.xlsxfor the Feb file, one prompt ("followmonthly-deck-instructions.mdfor February"), get a Feb deck in ~2 minutes. This is how 8 hours becomes an afternoon. - Share (5 min): One volunteer deck on screen. Facilitator asks only: "What would you still check by hand before sending this to leadership?" (Collect answers — feeds Segment 6.)
Facilitator notes: PowerPoint fidelity is the flakiest part (acknowledged on the Dana call) — set the expectation in minute one of this build: "First draft in 10 minutes, then you Tune. Budget your reps; slide-perfect takes iterations, and that's normal." If a pair's deck layout breaks, fall back to: Cowork outputs charts as images + a slide-by-slide content doc, pair assembles in the template manually. Still a win; don't burn 15 minutes debugging fonts.
3:00–3:20 — Trust, Checks & Guardrails (20 min)
Why this exists: Dana — "that will definitely be top priority… after we use AI to produce something that's going to be deliverable." Skip this and the team won't (and shouldn't) use any of it on real work.
Teach a 4-step checking routine (on the handout; pairs apply it to one output from each build):
- Trace one number to source rows (they've done this twice today — name it as a habit).
- Ask Claude to show its working: "List every cleaning step, filter, and assumption you applied. What did you exclude?" (Prompt 15). Read it. Disagree where needed.
- Cross-foot: totals in dashboard = totals in Excel = totals in deck. Prompt 16 does this comparison across their three outputs and reports mismatches. Run it live as a group.
- Human owns commentary & sign-off: numbers can be drafted by Claude; interpretation, commentary, and sending are a named person's job.
Guardrails (5 min, plain language): synthetic data today; real data only via their approved Snowflake/NetSuite path + IT policy; Cowork only sees the folder you point it at — make a clean working folder per task, don't point it at your whole drive; be extra careful with payroll/bursar (personal) data — check Brightmoor policy before any real use; if a number matters and you can't trace it, don't ship it.
3:20–3:30 — Close: Your Monday-Morning Task (10 min)
- Each person writes down one real, repetitive task they'll run this way in the next two weeks (own the sentence: "Every [week/month] I turn [file] into [output] for [audience]"). Pairs swap and sanity-check each other's pick against two questions (from the Nvalo framework, in finance language): Is it repetitive enough that even a rough first draft saves time? Can Claude see all the inputs it needs in one folder? If yes to both: start there.
- Recap on one screen — the only three things to remember: (1) Cowork acts on a folder and runs code — describe the output, then check it. (2) Talk → Jam → Make → Tune, one change at a time. (3) Trace every number you'd sign your name to.
- What happens next: prompt handout + data folder are theirs to keep; Snowflake connector removes the "export to Excel" step when IT provisions it (same prompts, new first line); suggest a 4-week show-and-tell where 2–3 people demo their Monday-morning task — that's the adoption engine for a busy team, not more training.
- Explicitly NOT homework: no Terminal, no GitHub, no "go explore AI." One task, done twice, beats ten experiments.
Overflow (3:30–4:00 hold): 1:1 help, Snowflake-connector questions with Dana/Greg, collect the Monday-morning tasks if Brightmoor wants a follow-up list for invoicing/follow-up session planning (headcount confirmation for Greg).
Appendices (facilitator reference)
A. Timing at a glance
| Time | Segment | Mode |
|---|---|---|
| 0:00–0:10 | Welcome: the 8-hour job | Facilitator |
| 0:10–0:25 | Cowork vs. chat (Excel clean-up demo) | Demo |
| 0:25–0:30 | Pair setup check | Hands-on |
| 0:30–1:20 | Build 1: Executive dashboard (HTML) | Hands-on |
| 1:20–1:25 | Break | — |
| 1:25–2:15 | Build 2: Analysis in Excel (reconciliation, variance, forecast) | Hands-on |
| 2:15–3:00 | Build 3: Monthly PowerPoint from template | Hands-on |
| 3:00–3:20 | Trust, checks & guardrails | Demo + apply |
| 3:20–3:30 | Close: Monday-morning task | Pairs |
| 3:30–4:00 | Buffer / 1:1 / Snowflake Qs | Optional |
Hands-on share: ~2h20m of 3h30m (~67%). Theory: ~15 min, all inside demos.
B. Troubleshooting (finance-flavoured ASK Ladder)
Teach this only when the first person gets stuck (project it then):
- Ask Claude to fix it — paste the error / describe what's wrong, in the same conversation.
- Simplify — "Explain what you just did like I'm a smart colleague who doesn't code. Where did it go wrong?" Then re-ask with that understanding.
- Kill & restart — fresh Cowork conversation, same folder, re-run from your saved instructions file. (In finance terms: like re-running a report from source rather than patching a broken spreadsheet.)
Common issues: wrong folder pointed at Cowork (re-point, don't reinstall) • file locked because it's open in Excel (close, re-run) • chart "not what I meant" (be specific: chart type + sort + units) • numbers differ between pairs (different cleaning choices — document them) • PowerPoint layout breaks (fallback: charts as images + content doc, assemble manually).
C. What was deliberately cut
- ASCII/location opener (replaced with the GL clean-up — same point, their data)
- Vacation planner / generic app builds, open-ended "build anything" time (this team wants workflows, not exploration)
- Terminal, GitHub, deployment/hosting, Plan Mode, agents/MCP roadmap (non-technical mandate; hosting = post-session IT question)
- Public-workshop showcase format (busy team; pair shares only)
- Live Snowflake as a dependency (connector is a bonus demo if IT provisions it; session must not hinge on it)
D. Open items to confirm with Brightmoor before Thursday
- Claude Cowork + Claude in Excel provisioned for all 20–30 attendees (largest risk — see Pre-work)
- Final headcount for invoicing (Greg); NDA if Brightmoor wants facilitators seeing real formats
- Snowflake connector: provisioned or not? (Either way the session runs; only Segment 8's optional demo changes)
- Can Dana share a scrubbed copy of the real monthly template? If yes, swap it for the synthetic one in Build 3 — highest-value single upgrade to this plan