--- title: "Your AI Is Guessing. Here’s What It’s Missing" publishedAt: "2026-10-14" dateModified: "2026-10-14" author: "Brian Fletcher" role: "Principal, Co-founder @ FM" headshot: "/images/brian_fletcher.jpg" category: "Thoughts" slug: "your-ai-is-guessing" excerpt: "Everyone is racing to put AI to work, but the missing ingredient isn't the model. It's context: what your business knows, how it works and what it's learned. Here's how to tell whether your systems actually have it." image: "/images/posts/your-ai-is-guessing.jpg" --- I've spent most of my career building software for businesses. The biggest problem is rarely the technology. It's what the technology doesn't know. Everyone is rushing to put AI to work. Assistants, agents, automations, custom GPTs. A lot of it is impressive. But there is a missing ingredient that gets far less attention than the models do. Context. It's the difference between AI that can do something and AI that knows how to do it for your business. ## What context means Think about hiring a great new employee. Smart, experienced, capable. On day one, they still don't know your customers, your processes, your priorities or the thousand small things that make your business tick. They have to learn. AI is no different. In practical terms, business context is five things: 1. **What you know.** Products, services, customers, pricing and company knowledge. 2. **How you work.** Processes, workflows, responsibilities and business rules. 3. **What you've learned.** Past projects, decisions, wins, losses and the lessons from each. 4. **What's happening now.** Sales activity, project status, financial performance, customer interactions. 5. **What matters most.** Goals, priorities, constraints and what success looks like. Without those, even the most sophisticated AI is making educated guesses. ## What context changes Ask AI to write a proposal for a prospective customer. Without context, you get a well-written, generic proposal. Give it your pricing guidelines, relevant case studies, past proposals, the customer conversations and the reasons you won or lost similar work, and now you have something you'd actually send. Or take customer service. An assistant can answer a question about your return policy. With that customer's purchase history, past complaints, service agreement and open issues, it can resolve the actual problem. Same model. Different outcome. The intelligence lives as much in what the model knows about your business as in the model itself. ## The problem I keep running into Most businesses already have most of this information. It's just everywhere. Your CRM knows your customers. Your accounting software knows your numbers. Your project tool knows what the team is working on. And your employees know everything that never made it into any of those systems, which is usually the most valuable part. None of it connects well. So people become the glue. They search, copy, paste, reconcile, explain and carry information from one place to another. I've watched businesses spend enormous amounts of time working around software that was supposed to make their lives easier. Now we're adding AI to that same disconnected environment. Are we making the business smarter, or just helping people work around its problems faster? ## How we build an agentic operating system for clients We start before we build anything. We sit down with your team and map how the business actually operates. Where information lives. Who knows what. Which systems need to talk. Which decisions keep getting made without the full picture. Then we build the layer that connects it. Shared memory that holds what the business knows. Connections into the systems you already run, whether that's your CRM, email, project tools or finance. And agents that work inside that context rather than beside it, so a proposal, a status report or a customer answer draws on everything the business has learned. We call that an agentic operating system. The first piece might be a single agent. It might be connecting two systems that already exist, or replacing a manual process with a simple automation. The point is to put intelligence where it changes the outcome, on a foundation that gets smarter with every project. And you own it. ## Three questions to ask about your business If you're exploring AI, start here. ### 1. What does your business know that your systems don't? Everything your people understand from experience that isn't documented or accessible to anyone else. ### 2. Where are people doing work the software should be doing? Repetitive handoffs, manual reporting, hunting for information, decisions that require five open tabs. ### 3. What would improve if your systems shared what they know? Better proposals. Faster service. Smarter staffing. More accurate forecasts. Fewer mistakes. Pick one meaningful opportunity and start there. You don't need to boil the ocean. ## The bigger picture We're moving from software that helps people complete tasks to software that helps businesses get better over time. Every customer interaction can inform the next one. Every project can make the next project better. Every decision can add to what the company knows. That only happens when you deliberately connect the knowledge, systems and workflows that run your business. It's why we talk about software that compounds. Faster work today is nice. Better work tomorrow is the point. Your people are learning every day. Make sure your business is learning with them. --- **Build your own agentic operating system** If any of those three questions hit close to home, this is the work we do. We've laid out what an agentic operating system is, what it connects and how we build one with you. [See how it works](/c/agentic-os).