Why AI Feels Generic at Work
You've tried using AI at work. The outputs were... fine. Technically correct. But somehow off. The email didn't sound like your company. The analysis didn't account for your industry's quirks. The recommendation ignored the unwritten rules that everyone in your organization knows but nobody has documented. AI feels generic because it is generic — it doesn't know your company, your clients, your jargon, or your culture.
Here's the secret: you can fix this in an afternoon. No engineering team. No enterprise software license. No technical background required. Just Claude, your company's key documents, and a few hours of setup.
Step 1: Write Your Company System Prompt (30 Minutes)
A system prompt is the "personality" you give AI before asking it anything. It's the difference between talking to a generic assistant and talking to someone who actually understands your business. Here's a template:
"You are an expert assistant for [Company Name]. We are a [industry] company based in [location]. Our main clients are [describe ideal clients]. Our tone is [professional/casual/technical/warm — however your company communicates]. When you write for us, avoid [things your company never says] and always [things your company always does]. Key terminology in our industry: [list 5-10 terms and their specific meaning in your context]."
This alone transforms AI output. Instead of generic business-speak, you get responses that sound like they came from someone who actually works at your company.
Step 2: Create Your Company Bible (1-2 Hours)
Gather the documents that define how your company works. Not everything — just the essential knowledge a new, very smart employee would need to be productive:
Must-haves: Company FAQ or "about us" (even internal version), product/service descriptions, brand voice guidelines (if they exist), key process documentation, common client questions and approved answers.
Nice-to-haves: Recent strategy documents, competitive landscape notes, client case studies, common email templates, industry-specific regulations or requirements.
Compile these into a single document. It doesn't need to be perfectly formatted — Claude handles messy text well. What matters is that the knowledge is there.
Step 3: Build for Specific Use Cases (1-2 Hours)
Don't try to build a general-purpose company AI. Build specific tools for specific tasks. Here's what works best:
Client email drafter: System prompt + company tone guide + common email scenarios. Feed it a rough draft or bullet points, get back a polished email that sounds like your company.
Proposal generator: System prompt + previous proposals + pricing structure + win/loss notes. Describe a new opportunity, get a first-draft proposal that follows your proven structure.
Internal report writer: System prompt + report templates + recent data. Paste raw data, get a formatted report that follows your company's reporting style.
New employee Q&A: System prompt + company bible + HR documentation. A new hire can ask any question about company processes and get an accurate answer — without bothering five different people.
Meeting prep assistant: System prompt + client history + industry context. Before a client meeting, paste the agenda and get a brief with background, likely questions, and talking points.
Step 4: Test, Iterate, Refine (Ongoing)
Run real work through your custom AI. The first outputs won't be perfect — they'll be much better than generic AI, but they'll still miss nuances. Keep notes on what it gets wrong. Every time you correct an output, add that correction to your system prompt or company bible. Over two weeks of daily use, your custom AI will go from "pretty good" to "eerily accurate."
The key insight: building a company AI isn't a one-time setup. It's a conversation you refine over time. Each correction teaches it something new about how your company thinks and communicates.
Why This Matters More Than You Think
A custom company AI isn't just a productivity tool. It's a knowledge management system. Every time you add a document, correct an output, or refine a prompt, you're creating an accessible, searchable repository of how your company works. New hires get up to speed faster. Institutional knowledge doesn't leave when people do. Best practices are encoded, not forgotten.
For small teams — especially in Southeast Asia where companies run lean — this is transformative. A 10-person company with a well-built company AI operates with the institutional knowledge of a 50-person company. It's not about headcount. It's about knowledge leverage.
Your Move
Open Claude right now. Write a system prompt using the template from Step 1. Add your company's FAQ. Ask it to draft a client email about a real situation. Compare the output to what you'd normally write. If it's close, you just built version 0.1 of your company AI. If it's off, note what's wrong — that's your improvement list for version 0.2. By next Friday, you'll have something your whole team can use.
The best company AI isn't the most expensive one. It's the one built by someone who actually understands the company.