How We Run Leaf Lane on AI Agents: An Honest Look at Our Stack and What We Have Learned
## Why This Post Exists
Most AI consultants write about AI in theory. We thought it was worth writing about how we actually use AI to run this company — not as a case study from a client, but from the inside.
This is not a pitch. It is an honest look at what works, what does not, and what we have learned running an AI-native small business.
---
## What "AI-Native" Actually Means for Us
Leaf Lane does not just advise on AI. Our operations run on it. Specifically:
- **Content is produced with AI agents.** Our blog pipeline runs on autonomous agents that research topics, draft posts, generate images, and publish — with human review before anything ships. The 34+ articles on this site were produced by that system.
- **Client intake is automated.** When someone submits the AI Quick Start Guide questionnaire, the data flows through a structured pipeline before a human ever reviews it.
- **Coordination happens through a multi-agent task system.** We use Paperclip to assign work, track progress, and maintain accountability across AI agents — the same way a traditional business would use a project management tool.
The result: a very small team can operate at a scale of output that would otherwise require 5–10 full-time people.
---
## The Stack
Here is what we actually use, and why:
### Content Pipeline
- **Agents**: Claude-based AI agents writing, editing, and publishing blog content
- **Image generation**: GPT-Image-1 via Supabase Edge Functions — each post gets a generated featured image automatically
- **Publishing**: Direct Supabase database writes; the blog pulls from Supabase on every request
### Client Operations
- **Forms and intake**: Next.js form wizard → Supabase → Stripe
- **Fulfillment tracking**: Order status managed in Supabase with a clear state machine (pending → in_progress → delivered)
- **Email**: Nodemailer via Netlify Functions for transactional messages
### Coordination
- **Paperclip**: Multi-agent task assignment, issue tracking, and governance. Agents check in, take tasks, report status, and escalate blockers — all via API.
- **Witness**: An epistemic coordination tool where agents contribute observations and decisions to shared knowledge boards. Useful for capturing judgment that does not fit neatly into a task.
### Infrastructure
- **Frontend**: Next.js deployed to Netlify
- **Backend**: Supabase (Postgres, storage, edge functions, auth)
- **Payments**: Stripe
---
## What Actually Works
**Speed.** A solo operator can publish at the pace of a content team. The bottleneck is not writing anymore — it is judgment about what to write.
**Consistency.** Agents do not have bad days. The newsletter, the blog queue, the intake process — these run on schedule regardless of whether I am available.
**Documentation.** Because agents communicate through structured task systems, there is a natural audit trail. You can always answer "what was done and why" for any task.
---
## What Does Not Work (Yet)
**Cold outreach.** AI can draft messages but cannot build genuine relationships. The first customers will come from human-to-human conversations. No automation replaces the founder posting on LinkedIn from their own account.
**Judgment at the edges.** When something novel happens — a client need that does not fit the standard flow, a market signal that requires strategic response — AI agents need a human in the loop. The CEO agent flags these; the human has to act on them.
**Speed of trust.** Potential clients cannot see the agents working. All they see is the output. Social proof, testimonials, and case studies still matter, and those require real customers first.
---
## The Honest Assessment
Running on AI agents is genuinely leveraged. But leverage is multiplicative — it multiplies whatever you put in. If the human is not distributing, the content does not reach anyone. If the human is not following up, leads do not convert.
The AI handles the predictable, repeatable work. The human handles the work that requires relationship, judgment, and trust.
That is exactly the model we help clients build for their own businesses.
---
## Want to See What This Looks Like for Your Business?
The [AI Quick Start Guide](https://leaflane.co/ai-quick-start-guide) is how we start most client engagements: a $250 structured review of your current operations, delivered in 2 business days, that maps your highest-ROI AI workflows based on your specific tools and pain points.
It is the same methodology we used to build this stack — applied to your business.