--- title: "Death by a Thousand Subscriptions: The AI Containment Problem" date: "2026-04-09T20:10:03.670562+00:00" number: 12 author: "Keith Eddleman" tags: ["enterprise-ai", "ai-sprawl", "shadow-ai", "leadership", "process"] --- Your controller just flagged 75 different AI charges on corporate credit cards this month. Five different slide generators. Three meeting transcribers. A dozen ChatGPT Plus subscriptions. And that's just what showed up on the Amex — God knows what's on personal cards getting expensed. Welcome to AI sprawl. The new shadow IT. And nobody's talking about it because everyone's too busy celebrating "AI adoption." The numbers back this up: [78% of AI users are bringing their own tools to work](https://www.microsoft.com/en-us/worklab/work-trend-index/) — Microsoft calls it "BYOAI." Meanwhile, [only 26% of companies](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai) have organization-wide policies governing how employees use generative AI. And [60% of employees](https://www.microsoft.com/en-us/worklab/work-trend-index/) are using AI tools that haven't been vetted or approved by IT. That's not adoption. That's a free-for-all. ## You Might As Well Have 50 Different Companies Here's what I'm seeing at my own company right now — and I guarantee it's happening at yours. One account executive has cobbled together three AI tools costing $180 a month. He's extremely efficient. The guy in the next home office has three completely different tools and a crappy system. There's no sharing between them. No visibility. No process. No way to say "this approach works better than that one" — because nobody's doing the same thing. If you have 50 salespeople, you might as well have 50 different companies. Each person is running their own entirely unique workflow with their own hand-picked AI tools. There's no measurement for that. There's no optimization for that. It's Johnny bar the door — every man for himself, get out your corporate credit card and go nuts. That's not AI adoption. That's chaos wearing a lanyard that says "innovation." ## The CTO's Impossible Position Last week an account executive came to me excited about a tool called Beautiful.ai. "Keith, it took my spreadsheet and turned it into this amazing presentation!" I smiled and said, "Yeah, that's awesome." I was being sarcastic. I know three other applications that do it as well or better. But what am I supposed to say? "Actually, I know four tools that are better, so stop using that one"? Now I'm the fun police. I've crushed someone's enthusiasm for doing exactly what we asked them to do — adopt AI. This is the CTO's dilemma: **you either crush their initiative or enable the chaos.** Neither option works. And here's what makes it excruciating: you can't blame them. Why would an account executive wait for the CTO to evaluate and approve an enterprise tool when, in 30 minutes and $49, they've got a deck that's better than anything the marketing department ever produced? Personalized to their client. Done before lunch. The enterprise procurement process cannot compete with that speed. Full stop. ## Stop Being a Tourist The problem isn't that people are using AI. The problem is they're using it like tourists — grabbing the nearest shiny thing, solving one problem for one moment for one client, then moving on to the next attraction. What they should be doing is thinking like architects. Not "what app solves my problem right now?" but "what does my ideal workflow look like, and what do I need to build it?" This is the same instinct behind promoting my developers to micro-CTOs. I didn't tell them which tools to use. I told them to stop thinking like coders and start thinking like product owners. The same shift needs to happen across the entire organization — with account executives, with marketing, with ops. **You're not telling people to stop using AI. You're telling them to stop using it selfishly.** ## A Framework That Actually Works Here's what I'd recommend. It starts with Covey, not with software. **Step 1: Begin with the end in mind.** Don't start with tools. Start with the question: what does your ideal day look like? If money were no object and you had a magic wand, what would your workflow be? Blue sky it. No constraints. This forces people to think about outcomes before they think about apps. **Step 2: Know yourself.** Which parts of your job do you enjoy? Which are you good at? Which don't need help? Now identify the bottlenecks — the boring, mundane, difficult, complicated parts that eat your time and energy. Don't automate what you're already great at. Subtract the friction. **Step 3: Cohort up.** This is the key move. Don't let individuals go shopping alone. Form a small cohort — the people who feel the pain, plus one technologist who can evaluate the plumbing. The account exec knows the bottleneck. The technologist knows whether the tool has an API, an MCP server, or whether you could build the damn thing internally in two days and save the organization thousands. Together they test 2-3 tools in a short sprint. What works gets elevated to other teams. What doesn't gets killed before it metastasizes into another line item on the corporate card. **The energy comes from the bottom. The architecture comes from the middle. The CTO sets the standard.** ## The Real Cost Isn't the Subscriptions The $49/month charges aren't what's killing you. It's the invisible cost: - **No institutional knowledge.** When your best AE leaves, his three-tool AI workflow walks out with him. - **No optimization.** You can't improve what you can't see. If 50 people have 50 different systems, you have zero data on what works. - **No leverage.** An enterprise license for one vetted tool costs less than 50 individual subscriptions to 50 different tools. But nobody's doing the math because nobody knows what everyone's using. - **No security review.** Every unvetted AI tool is a potential data leak. Your AE just pasted a client's revenue numbers into a free-tier AI app with no enterprise agreement. Cool. Remember — 60% of employees are using tools IT hasn't approved. That means 60% of your data governance is a suggestion. ## What I'd Do Monday Morning Start one cohort. Pick one workflow — ideally the most painful one, the one where people are already self-medicating with AI tools. Get 5-6 people in a room (or a Slack channel) with someone technical. Walk through the framework: 1. Design the ideal workflow 2. Identify the bottlenecks 3. Evaluate 2-3 tools together 4. Pick one. Kill the rest. 5. Roll it out to the wider team. Give it a month. That's your proof of concept for containment. It won't be perfect. But it's a system — and right now, you don't have one. ## What I'm Watching - **The companion problem:** This is the other side of Issue #11 — ["Who Owns AI Inside Your Company?"](https://www.theaiplaybook.com/issues/who-the-owns-ai-in-your-company). Ownership without containment is a title. Containment without ownership is bureaucracy. You need both. - **Shadow AI is the new shadow IT.** Except shadow IT was a server under someone's desk. Shadow AI is 150 people pasting proprietary data into 150 different tools you've never heard of. ## My Vibe Hustles - **[Essentialist.io](https://agents.essentialist.io)** — One platform that handles outbound campaigns, reply classification, knowledge base, and CRM advancement. Not five tools duct-taped together. One system, one API, one place to look. That's what containment looks like when it's built right. - **[Salesnado.com](https://salesnado.com)** — The anti-sprawl sales tool. One AI agent, 15 capabilities, all via email. No login, no dashboard, no "let me show you this cool new app." Just reply and it works. - **[The AI Playbook](https://www.theaiplaybook.com)** — This newsletter. A CTO's field notes on what's actually working. [12 issues and counting](https://www.theaiplaybook.com/issues). - **[BassFishing.World](https://bassfishing.world)** — A fishing spot marketplace I built because every photo on my phone has GPS coordinates in it. [See how it works](https://www.theaiplaybook.com/projects#bassfishing-world). ## What My Team Is Building at Q1Media Remember those micro-CTOs I mentioned in [Issue #10](https://www.theaiplaybook.com/issues/i-told-my-developers-to-never-write-another-line-of-code)? Here's what they shipped in the last two months using Claude Code: - **The BDR App** — Sits between our main Console and multiple CRM enrichment APIs and databases. It's the bridge between our ad tech operating system and every data source our business development team needs. No more alt-tabbing between six tools. - **The Events App** — Helps marketing and account executives calculate the precise ROI of trade shows and market events. Not a guess. Not a vibes-based recap. Actual numbers tied to pipeline. - **The Proposal Builder** — An agentic media proposal builder sitting on top of ten years of campaign data across thousands of campaigns and multiple ad exchanges. Proprietary data that nobody else has — now accessible through AI instead of buried in spreadsheets. Four real applications. Two months. That's what containment looks like when you stop letting everyone shop alone and start building together. [See everything we're working on](https://www.theaiplaybook.com/projects). ## The Bottom Line AI sprawl isn't a budget problem. It's a systems problem. And systems problems don't get solved by cutting credit cards or sending a memo. They get solved by giving people a better way to work than the one they cobbled together on their own. Contain the chaos. Keep the energy. Build the system. One cohort. One workflow. One month. Start there.