Winning work — sales
- Lead lists researched and personalised for outreach
- Job posts turned into tracked pipeline entries
- Proposals drafted, rendered and e-signed
Resources / Claude Code showcase
Free resource · A real working example
How we run an entire business with Claude Code
Flowlamina practises what it sells. Our own sales, client work, content and bookkeeping all run through one AI workspace — a tool called Claude Code. This page shows the actual setup, starting with the big picture and working down to the nuts and bolts.
The big idea
Forget the jargon for a moment. Every automated workflow on this page follows the same four-step shape:
You ask in plain English
"Draft a proposal for this job." "File this meeting." "Import this bank statement." No forms, no code.
The AI does the work
It reads your files, follows your written rules, and operates the tools you already use — email, CRM, spreadsheets, accounting.
Human checkpoint
You review & approve
Anything important stops and waits for a person. The AI does the labour; you keep the judgement.
Everything files itself
Documents saved, records updated, statuses tracked — the paperwork keeps itself, automatically.
One workspace, four parts of the business
This isn't one clever trick — the whole business runs out of the same AI workspace. Here's the map:
Serving clients — delivery
- Client folders and profiles kept automatically
- Meeting transcripts filed and summarised
- The builds themselves — chatbots, integrations, automations
One AI workspace
Claude Code + your business rules + your tools
Content & marketing — audience
- Videos edited, cleaned up and thumbnailed
- Descriptions written from the actual recording
- Social content pipelines for clients
Money & admin — back office
- Bank statements parsed straight into the books
- CRM and dashboard kept current overnight
- A daily 5am briefing on where to focus
If those four parts look familiar, they should — they're the same four quadrants we automate for clients. This page is simply our own business run through them.
A real example: proposals in minutes, not evenings
The clearest way to understand the setup is to watch one workflow. Writing a proposal used to take an evening — drafting, formatting, filing, updating the pipeline. Now it looks like this:
Paste the brief
A job ad, call notes, or a rough scope — whatever exists gets pasted into the chat.
The AI drafts it
It writes the full proposal in our voice, modelled on past winning proposals, and logs it in the sales pipeline.
Human checkpoint
Review, then ship
After a human read-through, one more command renders the branded PDF and preps it for e-signature.
❯ /generate-proposal
reading past winners + drafting rules…
✓ proposal drafted (scope, price, terms)
✓ pipeline updated → Drafting
paused for your review — edit, then:
❯ /finalise-proposal
✓ branded PDF rendered
✓ pipeline updated → Ready · e-sign prepared
For the curious — the full detail
The proposal pipeline, step by step
Every stage below runs today, exactly as described. The commands in green boxes are the actual "skills" — saved instructions the AI follows the same way every time.
1
Paste the brief
A job post, call notes, or a rough scope typed straight into chat — no template to fill in.
2
Draft
/generate-proposalThe AI reads past winning proposals and the house drafting rules — tone, structure, terms, Australian English — then writes the full proposal as an editable document and creates the tracking row in the sales pipeline.
pipeline status → DraftingReview & edit
Human gateThe system stops on purpose. Nothing client-facing leaves without a human reading it — scope, pricing and wording get adjusted before anything is rendered.
3
Finalise
/finalise-proposalRenders the accepted draft into a branded PDF — cover page, contents, page numbers, consistent tables — and preps it for e-signature upload.
pipeline status → ReadyClient signs
Human gateThe PDF goes out through an e-signature service. The client signs on their own time.
4
File the signed copy
/save-signed-proposalThe signed PDF is filed next to the original, the pipeline flips to Signed, and the engagement timeline gets a dated entry — the record keeps itself.
pipeline status → SignedThe pattern to notice: the AI does the drafting, rendering and record-keeping; the human owns judgement and sign-off. Every workflow on this page follows that split.
The full command library
Each chip below is a real command that runs today — a documented process the AI executes the same way every time.
Sales & outreach
Scraped lead lists get qualified, enriched with decision-maker emails, and personalised one by one with genuine web research. Job posts become tracked pipeline entries in seconds; campaign analytics sync back weekly.
/qualify-leads
/enrich-leads
/personalise-leads
/parse-upwork
/sync-instantly-analytics
Thousands of leads researched individually — a job no human would do by hand.
Delivery & engagement records
New clients get a scaffolded folder with a living profile. Meeting transcripts are filed and summarised into a running timeline, and the day's work is logged from a plain-language brain-dump. The AI also does the builds themselves through direct connections to client tools.
/client:create
/client:save-meeting
/log-day
Connected to email, calendars, CRMs, automation platforms and hosting — it operates the tools, not just talks about them.
Video & content engines
A full video post-production line: silence-cut and level a raw recording, scan the screen for accidentally-visible passwords and blur them, generate the thumbnail, and draft the description from a transcript of the actual narration.
/process-video
/redact-video
/generate-thumbnail
/generate-description
Once a process runs twice, it gets encoded — the third run is a single command.
Back office & strategy
Bank statement PDFs are parsed straight into the finance records — safely re-runnable, so nothing ever duplicates. A nightly sync keeps the CRM and a live dashboard current, and each morning an advisor briefing reads the whole business and returns a ranked list of where to focus.
/import-bank-statement-pdf
/sync-crm
/advisor
The advisor runs on a schedule at 5am — the business gets reviewed before the day starts.
Why it keeps getting better
The commands are only half the setup. The other half is a written memory: corrections and preferences get captured in the right file, so the same feedback never has to be given twice.
The operating manual. Hard rules, a map of the workspace, conventions — the first thing the AI reads every session, so behaviour stays consistent.
Read at the start of every sessionBusiness insights that repeat: what lifted reply rates, what clients actually prefer. Each entry states the rule, why, and when to apply it.
Written when a lesson generalisesA permanent log of structural choices — vendors, schemas, architecture — with the alternatives considered, so old analysis is never repeated.
Written when a meaningful choice is madeScoped rules for one workflow or one client: layout preferences, voice, process corrections. Loaded automatically whenever that work runs.
Written the moment feedback is givenBuilt to be trusted
Human gates where it matters
Anything client-facing — proposals, published videos, outbound email — stops for review before it ships. Automation handles the labour, not the judgement.
Secrets stay secret
Passwords and keys are referenced by name, never displayed or stored in documents. One command even scans screen recordings for visible keys before a video is published.
Fail loudly, never silently
Missing settings stop the run with a clear error instead of guessing at a default. Imports and syncs are safely re-runnable without duplicating data.
Want this for your business?
None of this needed custom software — it's an AI workspace pointed at processes you already run, with the judgement calls kept human. Tell us where your team loses time and we'll show you what your version looks like.
Book a free 30-minute consultation