🎮 The Next Input — Issue #202

In partnership with

⚡ The Briefing — 60 sec

• Claude Opus 5 became downright ruthless when tasked with running a vending machine Basically Opus 5 raising its fists at Sol 5.6 and saying, “I can be capitalist as fuck too.” Funny experiment, but also a useful reminder that agents optimise for the incentives you give them—even when those incentives produce a tiny vending-machine tyrant.

• Atlassian tightens tracking of staff AI use as other tech firms encourage “tokenmaxxing” Atlassian, you old fossil, you. While everyone else is trying to get employees using more AI, Atlassian appears to be standing near the token meter with a clipboard and a disappointed expression.

• AI labs buy, scan and shred millions of rare books Hide your Bitcoin and your OG copy of The Goblet of Fire. AI firms are going absolutely HAM on books lately. Apparently no corpus is safe once the model labs decide they need more tokens.

🛠️ The Playbook — AI Usage Control Tower

Mission
Create a transparent system that tracks AI usage, cost and outcomes without turning productive employees into suspects.

Difficulty
Intermediate

Build time
3–5 hours

ROI
Reduces wasted spend, identifies high-value AI workflows and replaces blunt usage restrictions with evidence-based governance.

0) Why This Matters

AI usage inside organisations is growing faster than most leaders can measure it.

Some employees are barely touching the tools.

Others are burning through tokens, building workflows and quietly producing three times as much work.

The wrong response is simply restricting usage.

The equally wrong response is letting every employee connect any model to any data source and hoping the invoice—and security review—works itself out later.

The useful middle ground is visibility.

Track what tools are being used, what they cost, what outcomes they produce and where the actual risks sit. Then govern based on evidence rather than vibes, panic or a monthly token leaderboard.

1) Architecture

2) Workflow

1. Inventory every approved AI platform, integration and internal agent.

2. Capture usage data by model, team, workflow and sensitivity level.

3. Link consumption to a named business outcome such as time saved, revenue influenced or cases completed.

4. Flag unusual cost, prohibited data access and unregistered tools for review.

5. Coach teams towards efficient models and prompts before imposing restrictions.

6. Review usage monthly and expand access where measurable value exceeds risk and cost.

3) Example Prompts

AI Usage Review

You are an AI operations analyst.

Review the supplied usage data.

Identify:
- highest-cost users and workflows
- highest-value workflows
- unusual usage spikes
- expensive tasks suitable for cheaper models
- teams with low adoption
- activity that may require governance review

Do not assume high usage is wasteful.

Separate:
1. productive heavy usage
2. inefficient usage
3. risky usage
4. unexplained usage

Recommend specific actions for each category.

Workflow ROI Assessment

Evaluate the following AI workflow.

Inputs:
- monthly model cost
- employee time previously required
- current completion time
- output volume
- error rate
- human review time
- business outcome

Calculate:
1. estimated monthly time saved
2. estimated operational value
3. cost per completed outcome
4. quality change
5. risk-adjusted ROI
6. scale, optimise or retire recommendation

State all assumptions clearly.

Token Efficiency Review

Review this AI workflow for unnecessary token consumption.

Identify:
- duplicated context
- oversized system prompts
- repeated document ingestion
- poor model selection
- avoidable retries
- output length beyond business need
- retrieval steps that can replace full-context prompting

Return a revised workflow that preserves quality while reducing cost.

4) Guardrails

• Measure business outcomes, not token volume alone.

• Tell employees what usage data is collected and why.

• Never use raw AI consumption as a standalone performance metric.

• Separate security investigations from routine productivity analysis.

• Use shared gateways instead of uncontrolled personal API keys.

• Apply stricter controls to sensitive data, not every workflow equally.

• Give teams an opportunity to explain anomalous usage before restricting access.

• Retain only the monitoring data needed for governance and optimisation.

5) Pilot Rollout — 3 hours

1. Select one team already using AI regularly.

2. Connect usage and cost data from its approved AI providers.

3. Register the team’s five most common AI workflows and expected outcomes.

4. Build a dashboard showing cost, volume, quality and estimated time saved.

5. Review one high-cost and one high-value workflow with the team.

6. Implement one optimisation and compare results after seven days.

6) Metrics

• AI cost per completed workflow

• Estimated hours saved

• Percentage of usage linked to registered workflows

• High-value versus unexplained token consumption

• Model cost by team and task

• Output error and rework rate

• Shadow AI incidents

• Cost reduction from model routing

• Employee adoption rate

• Governance exceptions per month

Pro Tip: Do not punish the person consuming the most tokens until you know whether they are wasting money or quietly carrying the department.

🎯 The Arsenal — Tools & Platforms

• Azure API Management · centralises AI access, metering and policy enforcement · Link

• Microsoft Entra ID · connects approved AI activity to organisational identity and roles · Link

• Power BI · visualises AI adoption, cost, risk and operational outcomes · Link

• Microsoft Purview · governs sensitive information and investigates risky usage · Link

• LangGraph · routes work across models according to cost, complexity and policy · Link

Copy-paste prompt block:

You are an enterprise AI usage and cost-governance architect.

Design an AI control tower for my organisation.

Organisation:
[DESCRIPTION]

Approved AI tools:
[LIST]

Known AI workflows:
[LIST]

Sensitive data categories:
[LIST]

Teams using AI:
[LIST]

Current reporting systems:
[LIST]

The system must:
- capture usage and spend by user, team, model and workflow
- connect AI consumption to business outcomes
- distinguish productive heavy usage from waste
- detect shadow AI and sensitive-data risks
- recommend cheaper models where appropriate
- avoid turning usage monitoring into employee surveillance
- provide transparent policies and review processes
- support role-based controls and auditability

Return:
1. architecture
2. usage data model
3. workflow registry schema
4. risk and anomaly rules
5. dashboard design
6. employee transparency policy
7. optimisation process
8. pilot rollout
9. operational metrics

💡 Free Office Hours

Most organisations will eventually need visibility into where their AI spend is going. The trick is building enough oversight to control cost and risk without discouraging the exact power users producing the strongest results.

Book here: https://calendly.com

Your prompts are leaving out 80% of what you're thinking.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

🕹️ Game Over

One lab is creating ruthless vending-machine capitalists.

Another is shredding rare books.

And your boss just wants to know why the token bill went up 38%.

— Aaron Automating the boring. Amplifying the brilliant.

Subscribe: link