Sources
Client files, meeting notes, proposals, emails, SOPs. Drop them in a shared folder. No formatting.
An operating system for your business.
"I know there's more I could be doing with AI. I just can't figure it out while running the business."
The real cost isn't time — it's inconsistency. No one holds the full state of every client, project, and commitment simultaneously. Things fall through not because you're careless, but because the system is manual.
The step change happens when AI maintains the compiled state of your business — and generates artifacts from that state.
Andrej Karpathy published this pattern: raw sources go in, an LLM compiles a living knowledge base, you query against it. I've adapted it for small businesses.
| Today | With Company OS | ||
|---|---|---|---|
| Knowledge | In your head + scattered docs | → | Living wiki — compiled, cross-referenced, linted |
| Tools | Claude as a chatbot, one-off prompts | → | Skills that read from compiled business state |
| Sources | You manually dig through email, docs, CRM | → | Team drops inputs into a shared folder |
| State | Reconstructed from memory every time | → | Persistent — updated as inputs arrive |
| Consistency | Depends on what you remembered to include | → | Every output reads from the same compiled truth |
| Onboarding | Hours walking someone through everything | → | /onboard generates a tailored brief in minutes |
"The tedious part of maintaining a knowledge base is not the reading or the thinking — it's the bookkeeping. LLMs don't abandon maintenance tasks; humans do."
The Company OS handles the bookkeeping — cross-referencing clients, updating project status, maintaining your SOPs — so you focus on the work that actually needs you.
Client files, meeting notes, proposals, emails, SOPs. Drop them in a shared folder. No formatting.
Claude updates living pages — one per client, per service, per team member. Cross-references and flags staleness.
Pre-built routines that read compiled state. Client updates, proposals, onboarding briefs, weekly reviews.
Your business rules. Brand voice, quality standards, naming conventions. Business rigor without policing.
| Source | Type | How ingested | Frequency |
|---|---|---|---|
| Client meeting notes | Drive scan | Checks shared folder for new or modified docs | Daily |
| Proposals & SOWs | Drive scan | New proposals auto-detected in client folders | On save |
| Email threads | Auto-pull | Gmail reads relevant client threads | Daily |
| CRM updates | Auto-pull | Pulls pipeline changes, deal stage updates | Daily |
| Project status docs | Drive scan | Team updates status docs; wiki absorbs changes | On save |
| SOPs & process docs | Ad-hoc | Dropped into shared folder by team member | On drop |
| Financial summaries | Ad-hoc | Monthly P&L or invoice data dropped in | Monthly |
Connectors pull from Gmail, CRM, and other tools directly. No human action.
Daily check of the shared folder for new or modified docs since last run.
Team member drops a doc into the folder. Picked up on next ingest cycle.
Runs on schedule — no manual trigger needed.
Email threads, CRM changes, calendar events for active clients.
Check shared folder for new or modified docs since last run.
Update client pages, project status, SOPs. Flag contradictions or staleness.
Record what changed. Surface anything that needs your attention.
| Automated output | When | What you get |
|---|---|---|
| Daily briefing | Daily after ingest | 3–5 bullet summary of what changed across clients |
| Monday review | Monday AM | Full /weekly-review output for your week ahead |
| Friday lint | Friday PM | /lint — stale pages, missing updates, contradictions |
Everything else (proposals, client updates, onboarding briefs) stays on-demand.
Write a detailed prompt. Provide all the context. Hope the output is good. Start from scratch next time.
Quality depends on the prompt. Every conversation is a blank slate.
The architecture creates the context before you type a word. Your business state loads automatically. The schema enforces your standards. Memory carries institutional knowledge.
Quality depends on the system, not the prompt. Every conversation starts with full business context.
A new team member with zero AI experience gets the same output quality — because the architecture does the heavy lifting.
Pull from email. Cross-reference the project doc. Check what was promised in the proposal. Write the update. Hope you didn't miss a thread.
30–45 min/client-update reads the client wiki page + recent activity → drafts update in your voice. You review, adjust, send.
Search email. Open the proposal. Check the SOW. Ask the team member who was on the call. Piece it together from three sources.
10–15 minThe client wiki page already has the compiled answer — scope, timeline, open items, last contact. Ask the OS and get the answer with citations.
InstantSpend two hours walking them through every client, every process, every tool. Repeat for the next hire. Half of it's out of date by next month.
2+ hours per person/onboard generates a tailored brief — all active clients, their role, your processes, your voice, open projects. Always current.
Open the last proposal. Copy the structure. Rewrite for this client. Check the pricing. Match the voice. Format.
60–90 min/proposal reads client wiki + your pricing + your voice → drafts a complete proposal. You customize the scope and adjust the framing.
| When | What happens | What you get |
|---|---|---|
| Monday | Sources ingested → /weekly-review | What changed across clients this week |
| Mid-week | Meeting notes + emails ingested | Client pages updated automatically |
| Friday | /lint | What's stale, what's missing, what needs attention |
| Monthly | /business-review | Health overview across all clients and operations |
The wiki absorbs information as it arrives. Skills produce artifacts on schedule. You stop doing the translation in between.
| Activity | Time today | With Company OS |
|---|---|---|
| Client status updates | 3–4 hrs/week | ~30 min |
| Proposal prep | 60–90 min each | ~15 min |
| Ad-hoc status questions | 2+ hrs/week | Instant |
| Employee onboarding | 2+ hrs per person | ~2 min |
Claude knows your voice and your team. Skills produce consistent outputs. Static context files.
Claude knows your business history. Skills read from compiled state. Living wiki that compounds over time.
Foundation is the system. Company OS is the memory.
The Company OS handles the bookkeeping — cross-referencing, status tracking, report drafting, consistency checking — so you focus on the work that actually moves the needle: client relationships, strategic decisions, and the judgment calls only you can make.
Let's grab coffee and talk about what Company OS looks like for your business.
james@nobo.works →