Enough Games — Your Company Needs an AI Strategy
Four parallel tracks for building a real AI strategy: governance, training, fast-track PoCs with reusable infrastructure, and company-wide integration. Not a vague deck — a structured plan.
Listen to under 2-min AI recap •TL;DR ↓
Enough Games — Your Company Needs an AI Strategy
Let me be specific: I'm talking about businesses that have a tech department. If that's you, you need an AI strategy. Not a vague "we should use AI more" slide deck — a real, structured plan.
I've been drafting one for myself and I want to share the framework here.
Four Tracks, One Strategy
The way I see it, a solid AI strategy breaks down into four parallel tracks. Each one serves a different purpose, and they all feed into each other.
Track 1: Governance and Documentation
This is your foundation. Company-wide docs, clear guidelines for AI use, and a security assessment where necessary. Before anyone touches a tool, you need to define the rules of engagement. What data can be shared with AI services? What can't? Where are the boundaries? Write it down.
Track 2: Communication and Training
How you roll things out matters as much as what you roll out. This track covers when and how you communicate changes, how you promote training, and how you build internal buy-in. People need to understand why this is happening, not just that it is.
Track 3: Fast-Track Initiatives and PoCs
This is where you identify quick wins — easily achievable goals that don't require building custom tools from scratch. Integration over invention. Something like Claude connected to n8n for workflow automation is a perfect example. Low effort, high signal.
But this track goes deeper than quick wins. It should also include building your reusable infrastructure: CI/CD pipelines, DevOps tooling, your company's auth/SSO layer, permissions, and a full-stack headless app boilerplate with API and MCP support baked in. Think of it as a launch pad — purpose-built for spinning up new AI integrations and internal tooling fast.
This boilerplate becomes a collection of fully documented guidelines, development patterns, clear boundaries, agent definitions (I'm referring to the Claude ecosystem here), and security-approved setups. The beauty of it? It enables non-developers to build PoCs and ship them securely, while everything remains maintainable by devs because they defined how systems should be built. Developers set the rails. Everyone else gets to ride them.
Track 4: Broad, Company-Wide AI Integration
This is the big picture. A central point of intelligence that connects across your entire organization. The question is: where does the experience start?
Maybe it's Claude with MCP connections to your relevant services, scoped by account permissions. Maybe it's embedded in Slack. Maybe it's somewhere else entirely. The architecture matters less than the principle: one cohesive layer of intelligence, not a dozen disconnected AI tools.
The next level is having your own agent sitting in the cloud — bound to your controls, not sending data externally, trained on your internal data. This is where MCP-enabled systems become the core concept. They make it straightforward to plug AI into your existing services without reinventing every integration from scratch.
The Vision Is Closer Than You Think
As an engineering leader with an extensive tech background, I see enormous potential in this space. The internet is already moving away from tedious manual processes. Even Google search has changed radically. We have intelligence within reach in so many places now.
Here's a small side-track to illustrate the point: I can't wait for the day I can use a single entry point, ask it to plan a holiday — hotels, car rentals, the whole thing — and get a solid result back for my final approval. Skip all the manual searching, comparing, and context-switching so I can focus on what actually matters. And AI can surface ideas you wouldn't have considered on your own.
Now bring that back to your company. Imagine your internal data, tasks, calendars, handbooks, legal docs, BI reports, observability dashboards — all accessible through one intelligent layer. You'd skip so many steps and eliminate so much manual work.
One Critical Caveat
You need to scope AI capabilities carefully. This isn't optional. We've already seen what happens when it goes wrong — AI should not get the ability to delete your business with a poorly scoped prompt. Guardrails aren't a nice-to-have. They're the whole point of Track 1 existing before anything else ships.
What does your company's AI strategy look like today — and are you building it with intention, or just hoping adoption happens on its own?
TL;DR
- Every company with a tech department needs a real AI strategy now — not a slide deck, but a structured plan with four parallel tracks.
- Governance and documentation come first: define what data AI can touch, set boundaries, and write it all down before anyone ships anything.
- Communication and training aren't afterthoughts — people need to understand why AI is being adopted, not just be told to use it.
- Build a reusable boilerplate and infrastructure layer so developers set the rails and non-developers can safely build and ship PoCs on top of them.
- The end game is a single intelligent layer over your entire organization — internal data, tasks, docs, dashboards — not a dozen disconnected AI tools.
- Guardrails aren't a nice-to-have; scoping AI capabilities is what prevents a poorly written prompt from deleting your business.
Stop experimenting randomly — build the governance, the infrastructure, and the vision as one deliberate strategy or don't bother at all.
Geddy
Senior Web Engineer / Lead
Engineering leadership • AI innovation • Product thinking. 20+ years of web engineering, from independent contractor to engineering leader. Passionate about developer experience and product engineering.