The Stack I Actually Run

People always ask about the tools. "What do you use? What's your stack?" As if the right combination of subscriptions is the secret.

It's not. But the question isn't entirely wrong, either. Your stack matters — just not in the way most people think. It matters because the architecture of how your tools connect determines whether AI is a toy or infrastructure. The specific tools are almost interchangeable. The patterns are everything.

So here's how I think about a solo builder's stack. Not a list of products. A set of layers that any serious one-person operation needs.

Layer 1: The Knowledge Layer

This is the foundation. Everything else builds on it.

Your knowledge layer is where all your accumulated expertise, client context, project history, and reference material lives in a form that AI can actually access and use. Without this, every AI interaction starts from zero. With it, your AI systems operate with the context of someone who's been working alongside you for months.

What this looks like in practice:

• A structured database or knowledge store that organizes information by topic, client, and project
• A way to capture learnings as they happen. Not at the end of the week when you've forgotten the details
• A retrieval system that surfaces relevant knowledge when you need it, without you having to remember where you stored it

The specific tool doesn't matter much. It could be a vector database, a well-organized Notion workspace, a custom system, or even a collection of well-structured markdown files. What matters is that knowledge goes in, stays organized, and comes out when it's relevant.

The most common mistake here is treating the knowledge layer as a filing cabinet. Dump stuff in and hope you'll find it later. Instead, treat it like a well-trained assistant's memory. Every piece of information should be tagged, categorized, and connected to related concepts.

Layer 2: The Workflow Layer

This is where AI stops being a chat interface and starts being infrastructure.

Your workflow layer defines the repeatable processes in your business and automates as much of each one as possible. The key insight: most knowledge work is 20% genuine thinking and 80% predictable process. The workflow layer handles the 80%.

Common workflows that every solo builder should consider automating:

• Intake processing: New client inquiry → extract key details → check against your capacity and expertise areas → draft response
• Research synthesis: Given a topic or question → gather relevant sources → summarize key findings → highlight gaps and contradictions
• Document generation: Given a template and context → produce a first draft → flag sections that need human judgment
• Communication management: Incoming messages → categorize by urgency and topic → draft responses for routine items → flag items that need your direct attention
• Reporting: Pull data from your systems → generate status reports, invoices, or analyses → queue for review

You don't need to automate all of these on day one. Start with the workflow that consumes the most time for the least cognitive value. For most people, that's some combination of email triage and document drafting.

Layer 3: The Quality Layer

This is the layer most people skip. It's the reason most AI implementations produce inconsistent results.

Your quality layer is the set of checks, validations, and review processes that ensure AI output meets your standards before it reaches a client or gets published. Without it, you're playing Russian roulette with your reputation.

What this includes:

• Verification gates: AI-generated facts, figures, and citations get checked against primary sources before being used
• Consistency checks: Outputs conform to your style, tone, and formatting standards
• Domain-specific validation: In regulated fields, AI output gets checked against current rules, regulations, and standards
• Human review checkpoints: Clearly defined points where you review and approve before anything goes out

The quality layer is what separates "I use AI" from "I run an AI-native practice." Anyone can generate a document with AI. The value is in guaranteeing that the document is accurate, appropriate, and ready for professional use. That guarantee is your reputation, and it's worth protecting systematically.

Layer 4: The Persistence Layer

The persistence layer is what makes your stack get better over time instead of staying static.

Every time you correct an AI output, that correction should be captured. Every time you discover a pattern in your work, it should be encoded. Every time a workflow fails, the failure mode should be documented so it doesn't recur.

This layer turns your daily work into accumulated intelligence. After six months of operation, your systems should be noticeably better than they were on day one. Not because the underlying AI models improved (though they might), but because your knowledge layer is richer, your workflows are more refined, and your quality checks catch more edge cases.

Without persistence, you're on a treadmill. With it, you're building a flywheel.

Layer 5: The Operations Layer

This is the mundane stuff that keeps a business running. Invoicing, scheduling, file management, backup, security. It's not exciting, but neglecting it is how solo builders end up losing important files or missing tax deadlines.

The good news: most operational tasks are extremely well-suited to automation. Calendar management, recurring invoicing, file organization, backup schedules — these are solved problems with existing tools. The AI angle here is minimal because the non-AI tools already work well.

Don't over-engineer this layer. Pick boring, reliable tools. Automate what you can with simple rules and integrations. Save the AI sophistication for the layers where it actually matters.

What's Not in the Stack

Notably absent: social media management, complex CRM systems, project management tools designed for teams, and anything that solves a problem you don't actually have.

Solo builders have a powerful advantage here: you can be ruthlessly minimal. You don't need Salesforce when a spreadsheet tracks your 50 clients perfectly well. You don't need a project management platform when your task list fits in a single file. The right amount of tooling is the minimum that keeps things running smoothly. Not one tool more.

Every tool you add is a tool you have to maintain, update, and integrate. Every subscription is a recurring cost that needs to justify itself monthly. The best stack is the smallest stack that gets the job done.

Assembling Your Own

If you're starting from scratch, here's the order I'd recommend:

1. Start with the knowledge layer. Begin capturing and organizing what you know. This takes time but it's the foundation everything else depends on.
2. Automate one workflow. Pick the most repetitive, time-consuming process in your business and build an AI-powered workflow around it. Get it working reliably before moving on.
3. Add quality checks. As soon as AI output is going to clients, build verification into the process. Start simple. A checklist, a review step, and formalize over time.
4. Build persistence. Start capturing corrections and patterns. This is the flywheel that makes everything else improve.
5. Keep operations boring. Use proven, simple tools for the business basics. Don't innovate here.

The whole process takes 2-3 months to get the basics running, and it never really stops evolving. That's fine. The stack isn't a product you ship. It's a system you tend. And the tending is what makes it yours.