Paul Surkau's deployment began with a wide operating brief and an important limitation: the full business context was not yet complete. RECAST built the departmental structure that could receive that context safely, while keeping the public claim at the level the evidence supports.
Evidence at publication: 12 configured agents and 33 bindings were observed on 30 May 2026, alongside a mapped German Discord workspace, gbrain pages, Notion helpers and local document tooling. The record verifies the built environment, not sustained adoption or a commercial outcome; this case is therefore marked built, adoption pending.
The company
The retained brief describes a German business workspace spanning daily management, scheduling, automation, content, email, creative work, research, web search, advisory work, ad copy, prompt engineering, skill creation, Notion and documents. It later expanded to acquisition, community sales and administration lanes.
Because the source does not preserve a fuller description of Paul's business model, this page does not invent one. The case is about the operating architecture that was delivered and the adoption evidence that remains to be established.
The constraint
Without routing, each category would become another separate chat, tool or manual process. Paul would remain responsible for deciding where a request belonged, re-explaining the context and moving the output into the next channel.
At the same time, incomplete business context meant the system could not responsibly be described as a fully trained autonomous workforce. The right first step was to build clear departmental lanes, visible behavior and safe execution paths that could be deepened as operating material arrived.
Why existing tools were not enough
Discord channels and Notion pages can organize work, but they do not make the work route itself. A generic assistant could answer in the wrong channel, disappear behind a failed session, query Notion inconsistently or reach for unsafe shell behavior when a deterministic helper was required.
RECAST hardened the workspace around visible replies, channel allowlists, session cleanup, model-routing stability, scoped memory and constrained wrappers. The value is not simply “12 bots.” It is a surface where each role has an address and a limited set of permitted actions.
Transformation map
Before
Management, content, research and acquisition requests → Paul selects a chat and supplies context → manual handoff
After - built operating architecture
German Discord channel → named department agent → memory or constrained helper → visible reply → operator review and adoption evidence
One workflow, end to end
The following route reflects the configured architecture; the source does not retain evidence of routine production use.
- Input: Paul posts a business request in a mapped management, content, research, acquisition or administration channel.
- Routing: The binding sends it to the corresponding specialist rather than a universal assistant.
- Context: The agent checks its role instructions and retrieves available gbrain context.
- Tool choice: A Notion query or document task goes through its constrained helper; creative and research paths use their separately permitted tooling.
- Production: The specialist prepares the requested analysis, draft, document or next action.
- Visibility: The answer returns to the designated Discord surface, making silent routing failures observable.
- Human gate: Paul supplies missing business context and approves consequential external action while adoption is proven lane by lane.
What changed
- Built: Twelve agent roles spanning main operations, content, creative, competitor intelligence, strategy, ads, prompt and skill work, acquisition, community sales and administration.
- Connected: Discord routing, gbrain pages, a Notion helper and local document wrappers were configured within the workspace.
- Tested: Visible Discord replies, channel routing, session resets, scoped active memory and model-stability rules were hardened during implementation.
- Runtime verified: Twelve agents and 33 bindings were visible on 30 May 2026.
- Actively used: Sustained operator activity is not retained in the source. The page therefore does not present the deployment as adopted at scale.
- Provider or client dependency remaining: Full business context, provider credentials and activation of scaffolded WhatsApp or Telegram paths remain separate dependencies.
- Financial result reported by client: None retained.
Proof
- Agent receipt: Twelve named departmental roles were configured.
- Routing receipt: 33 bindings were observed at verification.
- Workspace receipt: German channels covered management, customers, content, research, automation, logs, social, creative, advisory, ads, prompts and skills.
- Control receipt: Command-owner rules and channel allowlists constrained execution.
- Claim boundary: Initial business context was incomplete and no adoption or outcome ledger was retained.
Technology - revealed last
The deployment uses OpenClaw 2026.5.7, Discord, billing-proxy and OpenRouter routing work, Ollama, gbrain, a Notion helper, PDF/OCR tools, Higgsfield, Printing Press, Apify and Kie.ai. WhatsApp and Telegram were scaffolded rather than presented as live channels. The execution posture is allowlisted, with helper wrappers preferred over broad shell access.
Client quote
No source-backed verbatim client quote is published for this case. RECAST separates a delivery record from a testimonial and does not write the latter on the client's behalf.
Next transformation
See how Kevin Hucker's growth workspace separates six specialist lanes around shared venture context. It compares two built multi-agent deployments that still await stronger adoption evidence: one with six focused growth roles, the other with a broader twelve-role shell.
Get your AI blueprint
If the operating brief is broad but the business context is still being assembled, the honest move is to prove one department at a time. Book a free AI audit to choose that first lane and define its evidence bar.
Source record
- OPENCLAW-DEEP-CLIENT-CASE-STUDIES.pdf, page 7.
- RECAST_CASE_STUDIES_2026_UPDATED.pdf, page 14, document RA-CASES-001, version 2.0.
- RECAST_WEBSITE_MASTER_BLUEPRINT.md, sections 10.4, 11 and 58.