About
Ten years in marketing. Three building the AI that runs it.
Amber Lan has worked every part of marketing, then spent three years building the agents that do it. Flow AI is that system, opened up to other businesses. This page is the receipts.
I was the marketing team of one.
Flow AI exists because I needed it first.
Ten years in marketing, across every part of it. Positioning and category work. Search, back when that meant Google and now that it means ChatGPT too. Content calendars, brand voice, social, email, outbound, paid, events, analytics. Not one specialism: the whole function, in agencies and in house, for businesses from local operators to enterprise software.
Which means I know exactly what a marketer's week looks like. You are one person against a list that assumes five. The strategy sits in a document nobody opens because there is no time to execute it. The content calendar slips, then slips again. Leads arrive and nobody follows up for four days. You are not bad at your job. There is simply more work than hours, and every quarter the list gets longer.
Three years ago I started building my way out of it. Not prompts and not a tool subscription: real systems. Agents with the strategy written into them, content engines that hold a codified brand voice, outbound lanes that research before they send, search tooling that gets a business cited by AI assistants and not just ranked on Google.
Then I ran a whole marketing function on it. One marketer, no team, the full remit at an AI company: strategy, launches, content, thought leadership for three executives, events, and the reporting. The system did the production. I did the judgment. That is the only reason one person could hold it.
Flow AI is that system, opened up. Every business owner I meet has the same problem I had, with less time and no marketing background. So the agents that ran my function now run theirs.
双语营销工程师 · Practitioner. Systems in production. Receipts included.
Running marketing solo at an AI startup.
Runs the whole marketing function at an AI startup in Auckland, one person plus an agent stack.
Category and positioning
Message architecture, the claims register, and the go-to-market plan across two quarters and a product launch.
An agent-run content engine
Brief in, audited platform-native content out. Every artefact passes a codified voice gate before it ships.
The clean-room version on GitHubExecutive thought leadership
Three leadership voices on LinkedIn, quarterly calendars, every statistic verified against a source before publication.
Marketing infrastructure as software
An HTML-to-video rendering pipeline (Playwright + ffmpeg), a live benchmark kiosk app, and interactive GTM plans and dashboards the leadership team actually uses.
Event go-to-market end to end
Exhibitor operations, booth systems, and collateral for national tech events, run as production systems rather than one-off scrambles.
Every part of marketing, done personally.
Not a specialist who outsources the rest. This is why Flow can run ten marketing systems and know when each one is wrong.
Strategy and positioning
Category work, message architecture, claims registers, launch plans across quarters.
Search and AI search
Technical SEO through to answer engine optimisation, with the toolkit published open source.
Content and brand voice
Calendars, long form, executive ghostwriting, and voice codified strictly enough that a machine can hold it.
Social and community
Platform-native distribution, founder voices, and the unglamorous work of replying on time.
Outbound and lifecycle
Researched prospect lists, sequences inside safe limits, email and follow-up that a human still signs off.
Paid, analytics and events
Ad creative and audiences, tracking wired properly, and national events run as systems rather than scrambles.
Proof
Shipped systems, linked to source.
Four featured, six more. Every one opens on GitHub.
geo-seo-claude
A GEO/AEO-first SEO toolkit for Claude Code: citability scoring, AI-crawler analysis, schema markup, PDF reports. Built so a brand gets cited by ChatGPT, Claude and Perplexity, not just ranked on Google.
View repository Content system · React + TSFlow AI
Automated, AI-search-optimised content for NZ service businesses. A short brief in, publish-ready service, location and FAQ pages out.
View repository Content engine · Python + ClaudeContent engine
One source document in, a month of on-brand posts out. Every line passes a codified voice audit, so the output reads human, not generated.
View repository Distribution · TypeScriptSocial Media OS
Plan once, adapt per platform, ship on schedule. The distribution half of the content engine.
View repositoryMore systems
Strategy through code.
Six capabilities, one person.
Strategy
Go-to-market, category positioning, message architecture, launch planning.
Demand
SEO and AEO/GEO, content engines, inbound diagnostics, outbound automation.
AI systems
Claude agents and skills, brand-voice codification, LLM pipelines, MCP integrations.
Engineering
Python, TypeScript, Swift, React/Next.js, Playwright + ffmpeg rendering.
Ship
Vercel, GitHub, CI, e2e testing. Production, not prototypes.
Bilingual
English and 中文, native Mandarin. Strategy and content in both markets.
How I work.
Four rules, applied to every system on this page.
Report, don't prescribe.
"I tested it, here's what happened" beats "you should." Observations from a real system, not advice from a podium.
Receipts persuade.
Numbers and screenshots, a before and an after. No receipt, no ship.
Failures are episodes.
What broke, the root cause, the fix, and what it cost. Published at the same rate as the wins.
Leverage over labor.
Anything done twice by hand becomes a system. The agent stack is the team; I stay on strategy and judgment.
One person with an agent stack, instead of an agency.
You talk to the person who does the work. No account manager, no junior doing the writing, no retainer for hours nobody can show you. Tell Flow what you sell and where growth is stuck, and a written read comes back within the week.