Z Media's initial scope was unusually broad: CEO coordination, knowledge ingestion, specialist agents, meeting follow-up, affiliate qualification, forecasts, decks and UGC support. RECAST turned that list into one governed internal operating layer, then audited what was live and what still depended on Remy's material or platform access.
Evidence at publication: a 2 July 2026 live audit found eight configured agents mapped to real Slack channels, verified Slack posting, active daily backups and a 173-file knowledge-base structure. Remaining dependencies are limited and named; no agency-performance figure is inferred from the deployment.
The company
Remy Beaumont runs Z Media, a TikTok Shop-focused agency spanning e-commerce content, affiliate programs, paid advertising, live shopping, client delivery and adjacent business opportunities.
The company has many potential leverage points, but each has distinct data and judgment requirements. Affiliate qualification is not finance. Creative output is not operations. A sales forecast is not a CEO briefing. The agency needed specialist lanes without forcing Remy to manually reconcile every lane.
The constraint
Too much operating work depended on manual Slack triage, Fathom and Linear context, disciplined SOP use and Remy's own bandwidth. The breadth of the requested system created another risk: a premium scope could look impressive while remaining vague about which modules truly worked.
The constraint was therefore both capacity and proof. Z Media needed an operating layer that could route real work and expose the boundary between a live module, a ready module and a client-dependent module.
Why existing tools were not enough
Slack carried conversation. Fathom captured meetings. Linear held tasks. Drive held knowledge. Those products did not automatically turn a call into a reviewable task draft, make the correct specialist own it, attach the relevant SOP or preserve a controlled record.
The missing layer needed to connect these systems without allowing a meeting transcript to create unreviewed production tasks or a specialist to operate outside its lane. For consequential work, “draft first” is the useful form of automation.
Transformation map
Before
Slack + calls + Drive + Linear + platform data -> Remy triages context -> specialist follows up -> manual record
After
CEO Direct or mapped Slack channel -> specialist route -> structured knowledge/tool path -> draft or governed output -> human review -> system record
The same model supports sales, finance, affiliate, operations, creative and research without pretending those departments are interchangeable.
One workflow, end to end
- Input: A meeting enters through the Fathom pull path with its transcript and account context.
- Extraction: The system identifies candidate actions and supporting knowledge.
- Routing: The relevant specialist lane receives the work according to its real Slack channel mapping.
- Draft: A Linear item is prepared as a draft rather than silently filed as an approved commitment.
- Human gate: Remy or the responsible operator checks assignment, wording, due state and client implications.
- Record: The approved task enters the normal operating system with its source context intact.
- Continuation: The knowledge base and channel history support later forecasts, decks, affiliate analysis or creative work.
Other modules follow the same pattern: prepare a usable asset, surface it in the team's working environment and retain the approval boundary.
What changed
- Built: A TikTok Shop OS with CEO Direct and specialist sales, finance, affiliate, operations, creative and research lanes.
- Connected: Slack socket routing, Drive-to-knowledge ingestion, Fathom pull, Linear draft filing and Sheets output paths.
- Operationalized: Affiliate qualification scripts, forecast-to-XLSX generation, PPTX deck creation and UGC pipeline support.
- Protected: Daily backup tarballs and a failover service reduce dependence on one fragile runtime state.
- Verified live: The July 2 audit found keys present, integrations running, Slack posting proven and agent routing working.
- Actively used: Live routing and posting are evidenced. The source does not quantify campaigns completed, affiliate revenue or time saved.
- Client dependency remaining: Deeper SOPs, other-shop channel confirmation, brand-kit material and TikTok Seller/Affiliate Center access for metrics outside the current API path.
Proof
- Agent receipt: Eight configured agents mapped to real Slack channels.
- Knowledge receipt: A 173-file knowledge-base structure with Drive sync and document structuring.
- Workflow receipt: Fathom meeting pull and draft-first Linear filing were part of the inspected system.
- Output receipt: Affiliate qualification, spreadsheet forecasts, presentation decks and UGC pipeline support were built.
- Resilience receipt: Daily backups and a failover service were active.
- Boundary: The audit classed modules as live, ready or explicitly access-dependent; it did not report a financial outcome.
Technology - revealed last
The system uses an OpenClaw gateway on a RECAST VPS, Slack socket connections, Claude Max through Claude Code subscription routing, an OpenClaw billing proxy, Drive sync, an Obsidian-style knowledge structure, Fathom, Linear, Google Sheets, XLSX and PPTX generation, ElevenLabs-supported UGC paths, backups and failover. The architecture matters because every tool sits behind a named operating lane and evidence state.
Client quote
No source-backed verbatim client quote is published for this case. The July audit and runtime artifacts carry the public proof.
Next transformation
See how Yannick Schatz's agency layer was hardened through real client delivery, including channel, memory, transcription and long-render failures.
Get your AI blueprint
If your agency has the tools but one person still turns every conversation into every next action, the bottleneck is the operating layer between them. Book a free AI audit to identify the first route worth proving.
Source record
OPENCLAW-DEEP-CLIENT-CASE-STUDIES.pdf, page 23.RECAST_CASE_STUDIES_2026_UPDATED.pdf, page 19, documentRA-CASES-001, version 2.0.RECAST_WEBSITE_MASTER_BLUEPRINT.md, sections 10.5, 11 and 58.