One Tool. Three Cancelled Hires.

Three weeks ago, I showed a client what Claude can actually do.. By Friday, they’d cancelled three junior analyst hires. Not because they panicked. Because the maths stopped making sense. Three analysts at $65K each is $195K a year before super, before onboarding, before the six months it takes to get them productive. Claude Pro costs $30 a month. The gap between those numbers killed the job ads. The part I don’t talk about much - eighteen months ago, I was telling clients to wait. “The tools aren’t ready, give it another cycle.” What changed my thinking? I cost at least two organisations six months of competitive advantage because the trend wasn’t clear enough for me to back it, but now - we are seeing stock market crashes and three cancelled hires in one week. If you’re reading this thinking you should’ve moved faster, you’re not alone. Most leaders I work with are outwardly positive on AI and privately terrified they’ll be the one who gets it wrong.

Today’s content is for you.

• The economics of Claude’s product line in CFO maths
• What $285 billion in stock market losses tells you about your competitive position
• Why “surface users” are burning your AI budget while “superusers” run 3x to 5x ahead
• 5 context-engineered tests to run Claude against your actual work this week
• The governance question your board should be asking by Monday Most Australian organisations already use AI - the problem is that most still can’t measure what it’s actually replacing or the ledger impact.

---

Before Claude → Junior analyst researches, formats, drafts. $65K/year. After Claude → Cowork reads your files, delivers formatted output. $30/month. Before Claude → Legal review: external counsel at $800/hour. After Claude → Legal plugin flags risky clauses, suggests alternatives. Included. Before Claude → Weekly competitive briefing: 2 days of analyst time. After Claude → Scheduled Cowork task: runs Monday 7am, briefing by 7:15am. Before Claude → Excel modelling: finance team, 4 to 6 hours per model. After Claude → Cowork builds the model from a 3-line brief, formulas included.

1. The $285 Billion Warning

In Feb, after Anthropic released Claude’s plugin system, software stocks lost $285 billion in a single day (Fortune, 3 February 2026). Thomson Reuters dropped 16%, its worst single-day loss on record. LegalZoom fell 20% (Bloomberg, 3 February 2026). This is institutional investors pricing in what happens when a $30/month subscription starts doing the work of $800/hour professionals. I watched it happen from my desk in Melbourne. Most Australian execs, I talk to had no idea it even occurred. The plugins are the trigger. Sales. Marketing. Legal. Finance. Data analysis. Product management. Customer support. Each one turns Claude into your domain specialist. Install the Legal plugin and it reads contracts, flags risk, explains clauses in plain English, and suggests alternative language. The market priced in the disruption before boards even knew the product existed. Takeaway: If your board hasn’t discussed what a $30/month AI subscription means for your cost structure, make your you do on Monday!

2. Claude Isn’t a Chatbot. It’s 8 Products.

Executives still think Claude is “like ChatGPT but from Anthropic.” A text box. Yes - in 2024. In 2026, Claude is:

3. The Superuser Gap

Companies where employees are AI “superusers” generate 3x to 5x more revenue per head than companies with “surface users.” That gap is the biggest predictor of whether your AI spend produces returns or another line item. A surface user opens Claude, rewrites an email, closes the tab. A superuser builds a folder of context files. Teaches Claude their processes. Installs plugins for their function. Runs entire workflows through it. I saw this with the client who cancelled those three hires. They had 200 people using AI. Every day. When I asked what they were using it for, the answer was email rewrites and meeting summaries. Cool - but that is not AI adoption, it’s closer to $200K of annual subscriptions, with zero structural savings. You need - architecture. The folder structure. The context files. The plugins. The skills. One well-written context file about your role, your preferences, and your standards is worth more than 50 clever prompts. Ask your ops lead this week: how many of our AI users have built a single context file? I’m confident that within 18 months, organisations that haven’t moved teams from surface users to superusers will be paying 3x more per unit of output than their competitors. Which leads to a technology gap, then cost gap….and cost gaps kill businesses.

4. The Five Tests

I don’t write prompts anymore. I write context files. And I let the AI prompt me. The best results don’t come from clever one-liners you paste into a chat box. They come from giving Claude your actual context, your actual files, your actual constraints. Then letting it interview you before it starts working. Every test below uses this pattern. You give Claude a folder of context. It asks you questions. Then it builds the output. The more honest your answers, the sharper the result. Run these this week. Each one takes under 15 minutes. They escalate. The first is easy. The last one will make you uncomfortable.

Test 1: The Research Replacement

☐ COPY AND PASTE THIS ENTIRE PROMPT Create a folder on your desktop called "Weekly Brief." Drop in your last board report, your team's current priorities doc, and any strategy notes from this quarter. Open Claude Cowork. Add the folder. Then paste this: "Read everything in this folder. Based on my organisation's priorities, build me a competitive briefing on [your industry] covering the last 7 days. Use at least 5 sources. Show timestamps, note contradictions between sources, and cite each with working links. Format it so I could paste this into a Monday morning leadership email."Replaces: The junior analyst compiling your weekly brief. ~$15K/year of analyst time on this task alone.

Test 2: The Document Drafter

☐ COPY AND PASTE THIS ENTIRE PROMPT Create a folder called "Drafting Context." Drop in: your last 3 documents of this type (board papers, strategy memos, proposals), your org chart, and any brand or style guidelines you have. Open Claude Cowork. Add the folder. Then paste this: "Read everything in this folder. Learn my voice, my structure, and my audience. Before you write anything, interview me. Ask me 5 questions about what this [board paper / strategy memo / client proposal] needs to cover, who's reading it, and what outcome I want. Don't guess. Don't start until you've asked."It reads your past work, learns your patterns, then asks the questions a good chief of staff would ask before drafting. The output sounds like you wrote it because it started with your context, not a blank page. Replaces: 4 to 6 hours of drafting time before your first review pass.

Test 3: The Financial Model

☐ COPY AND PASTE THIS ENTIRE PROMPT Create a folder called "Finance Context." Drop in: your last 2 financial models or budget spreadsheets, your chart of accounts, and any assumptions docs. Open Claude Cowork. Add the folder. Then paste this: "Read my existing models. Learn my formatting, my assumptions structure, and my currency conventions. Now build a [describe the model, e.g. 12-month P&L projection] using Excel formulas (SUM, SUMIF, IF, VLOOKUP). Never hardcode calculations. Put assumptions in labelled cells. Format currency as $. Before you start, ask me 3 questions about the key assumptions I want to test."Replaces: The finance analyst building your model from scratch.

Test 4: The Process Encoder

☐ COPY AND PASTE THIS ENTIRE PROMPT Create a folder called "Process Context." Drop in: any SOPs, checklists, email templates, or example outputs from a process you repeat regularly. Open Claude Cowork. Add the folder. Then paste this: "Read everything in this folder. You're going to turn this into a reusable Claude Skill. But first, interview me. Ask me questions about the full workflow: what triggers it, what inputs I need, what the output looks like when it's done well, and what quality criteria matter. Build the SKILL.md file only after you understand the complete process."Replaces: The institutional knowledge that walks out the door when someone leaves.

Test 5: The Uncomfortable One

☐ COPY AND PASTE THIS ENTIRE PROMPT Create a folder called "Org Context." Drop in: your org chart, your current headcount and salary bands, your last 2 quarterly reports, and your AI initiative list. Open Claude Cowork. Add the folder. Then paste this: "Read everything in this folder. Before you answer, ask me 3 clarifying questions about which functions I'm most concerned about and where I think we're most exposed. Then be brutally honest: which functions in my organisation are most exposed to restructuring in the next 12 months? Show specific tasks, approximate FTE cost at Australian salary benchmarks, and what a Claude-based alternative looks like. Format as a table I could put in front of my CFO on Monday."Replaces: Nothing. This is the question nobody on your team will ask out loud. That’s why it matters.

Takeaway: If you ran all five without finding at least one hire, subscription, or process you’d change this quarter, you weren’t honest with the inputs.

How to Use These

• Download Claude desktop app from claude.ai/download ($30/month Pro plan)
• Build a context folder for each test. Drop in real files: strategy docs, org chart, past reports, financial models
• In Cowork, click “Add Folder” and select your folder
• Select Opus 4.6 + Extended Thinking (the smartest model, thinks before responding)
• Copy each prompt above. Let Claude read your context, then answer its questions honestly
• Run one per day this week. Time each one. Compare to your current process
• The gap - “what this took before” and “what it takes now” is your FTE case The difference between a surface user and a superuser is the folder. The context file does the work.

---

Before You Restructure Anything

Claude has read/write access to whatever folder you share. That includes files containing personal information under the Privacy Act 1988. ✦ Privacy Act compliance. If Claude processes personal information, you need to meet APP 11 (security of personal information). A consumer subscription doesn’t meet enterprise security requirements. Anthropic offers Team and Enterprise plans with different data handling. ✦ VAISS alignment. The Voluntary AI Safety Standard sets expectations for responsible AI use. “Voluntary” today. Regulatory sentiment is moving. ✦ Automated decision-making. If Claude’s outputs feed decisions affecting people, the OAIC’s guidance applies. Transparency isn’t optional. This isn’t a reason to wait- move with governance built in from day one.

Before Friday: Run Test 1 and Test 5. Takes 25 minutes combined. The research replacement shows the speed. The uncomfortable one shows the exposure. Not sure where your organisation stands? Take the 2-minute AI Readiness Scorecard: https://www.appliedaiaustralia.com.au/ai-tools/readiness If you got to the end of this, forward it to someone who should’ve read it last month. Ramon.

Catch Up - Applied AI Australia

In conversation with Andrew McCarthy GM, Notion. We discussed how Heidi Health used AI to eliminate 260 hours per month. ▶️ Apple Podcasts | Spotify

---

Subscribe to Applied AI Australia

All you need to know in one place, specifically for Australian Executives and Organisational leaders. Frameworks you can apply in 48 hours.

About

We specialise in translating AI complexity into strategy that drives business outcomes, Australian executives can act on with confidence. We deliver executable frameworks, board-ready playbooks, and strategy designed for Australian enterprises.Not sure where your organisation stands?

---