Case studies
The systems we eat our own dog food on.
Real client case studies start landing late 2026 after the first pilots ship. Until then — here are two internal R&D systems running on the same patterns we sell. Both started as our own pain.
Case 01
Admin Agent
Founder admin automation. Internal build. In production.
Pattern
Scattered admin tasks → one local agent → a single, signed-off queue.
The pain
A solo-operator founder loses 8–12 hours a week to admin scatter — email triage, calendar coordination, follow-up reminders, expense reconciliation, content drafts, social-media engagement queue. Each task is individually low-leverage; together they eat half a working day. Outsourcing to a VA adds context handoffs, mistakes, and privacy concerns.
The solution
A local agent with read access to email and calendar. Drafts replies, suggests calendar slots, surfaces stale follow-ups, queues expense entries. Founder reviews and sends — the agent never auto-sends to the outside world. All client / customer PII stays on the founder's machine; no third-party data sharing.
What we got back
- · About six hours per week back, mostly in the morning admin block
- · Sub-A$50/month run cost (local model + minimal API calls)
- · Zero client data left the founder's machine
- · Two months later: the agent caught a follow-up that turned into a paid engagement
Case 02
因果墨 · Causation Ink
Founder-led content production pipeline. Internal build. Running.
Pattern
Topic → image agent → video synth → staggered publish across YouTube, web, and LinkedIn.
The pain
Founder-led businesses need consistent content to build inbound. Production takes 8–15 hours a week. Most founders publish two pieces sporadically and then stop. Generic "AI writes for you" tools produce summary-of-someone-else's-work content that nobody reads. Real content needs first-hand evidence, angle work, and production polish — and most days the founder doesn't have time for any of it.
The solution
A multi-agent content pipeline: founder picks the topic (this stays human, on purpose), agents handle narrative structure, image generation, video synthesis with TTS, staggered publish across YouTube, web article, and LinkedIn post. The founder reviews about 30 minutes per pillar and signs off. The system never publishes without an explicit go.
What we got back
- · About 95% of production work removed from the founder's plate
- · Stable output: roughly 10–12 pillars per month → ~50 assets per month
- · Three-channel reach without the cadence collapse most solo operators hit
- · "First system breaks on day two" applied to ourselves: the v1 pipeline did break on day two, the v3 is what runs now
What this means for you
Both these systems use the patterns we sell. The engineering is the same.
We didn't build a different stack for ourselves and a "client version" for you. Admin Agent and 因果墨 use the same PII-isolation architecture, the same v1→v2→v3 iteration discipline, and the same human-signoff gate that goes into every engagement we ship.
If you're considering a pilot, you can read the production logic of these two systems as our open-book reference. We'd rather you see what we actually do than read a sales narrative about it.
Eight questions. Email reply in two business days.