Four routed agents separate command, content, sales and curriculum work inside the communication surfaces the business already uses.

Command work, public content, prospecting and curriculum development have different inputs and standards. Warren's sales-training operation needed those modes separated before the system could credibly behave like a team—and before anyone pretended it knew business context that had not yet been supplied.

Evidence at publication: four agents and 24 bindings were verified on 30 May 2026 across a hardened Slack and Discord deployment. The source also calls the available business context sparse, so the case is published as an operating shell with adoption evidence pending.

The company

Warren Mulvey's setup is centered on remote sales training and community. Its recurring work spans daily coordination, marketing and brand, sales outreach, learning material and the operational spaces that support a course or member community.

Those functions are related, but they are not interchangeable. A curriculum outline should not inherit the posture of a cold-outreach message. A daily operations request should not be buried inside a content brainstorm. The structure needed to reflect the business before more provider automation was added.

The constraint

Slack and Discord were present, but communication surfaces alone do not create departments. Without role and channel routing, a general assistant would continue to blur content, sales and training work, leaving Warren to classify requests and carry context between them.

The second constraint was evidence quality. The retained source does not provide detailed course materials, workflow volumes or completed outcomes. That limits what can responsibly be claimed. The first transformation was therefore a secure, extensible operating shell rather than a fictional story of full autonomy.

Why existing tools were not enough

Slack could host team conversation and Discord could organize working channels, but neither established which AI role should answer, what tool access it should receive or which context should remain specific to a department.

An AI team also expands the security surface. Broad shell access would give routine content or curriculum tasks unnecessary system power. The operating layer needed allowlisted execution, a hardened server and explicit channel-to-agent bindings.

Transformation map

Before

Slack + Discord -> mixed requests -> Warren identifies mode and supplies context -> work begins

After - built foundation

Department channel -> assigned agent -> allowlisted capability -> human-reviewed output -> department record

The architecture creates somewhere for future knowledge and workflows to live without claiming they have already been adopted at scale.

One workflow, end to end

The following is the configured curriculum route; routine production use is not asserted by the source.

1. Input: Warren places a training-module request in oc-training rather than the general command channel.
2. Routing: The binding directs it to the training/curriculum agent, keeping sales and content modes separate.
3. Context: The agent uses the business information available in its workspace and identifies missing source material instead of inventing a program detail.
4. Production: It prepares a structured lesson, exercise or revision for review.
5. Control: Any external publishing or member-facing action remains with Warren or the team; execution is constrained by the allowlisted posture.
6. Output: The draft returns in the correct communication surface.
7. Learning boundary: Approved curriculum source material can deepen that lane over time without automatically changing the sales or marketing agents.

What changed

• Built: Main/Warren AI, content/marketing, sales/outreach and training/curriculum agents, each with a dedicated operating lane.
• Connected: Slack and Discord were configured with command, content, sales and training channel mappings.
• Tested: Agent bindings, channel structure, key-only server access, firewall posture and allowlisted execution were established and inspected.
• Verified: Four agents and 24 bindings were recorded on 30 May 2026.
• Actively used: The source does not retain recurring task volume, member-facing outputs or a sustained adoption record. Those claims are deliberately withheld.
• Remaining dependencies: Richer curriculum, offer, brand and community context must be supplied; individual provider workflows then need separate connection and evidence.
• Financial result reported by client: None retained.

Proof

• Agent receipt: Four functional agents for command, content, sales and training.
• Routing receipt: 24 bindings at the live verification point.
• Workspace receipt: Discord includes daily operations, marketing, brand, strategy, workspace, prompt-builder and skill-builder areas.
• Security receipt: Key-only SSH, UFW and allowlisted execution reduce unnecessary access.
• Claim boundary: The evidence supports a flexible operating shell, not a replacement headcount, revenue result or completed curriculum volume.

Technology - revealed last

The deployment runs OpenClaw 2026.4.26 on a Hostinger VPS with Slack, Discord, local billing-proxy routing and OpenRouter fallback. The server uses key-only SSH, UFW and an allowlisted execution posture. Channel bindings map the four operating roles across oc-command, oc-content, oc-sales and oc-training, with a broader Discord workspace prepared for daily operations, brand, strategy and system development.

Client quote

No source-backed verbatim client quote is published for this case. The public story is limited to the documented deployment; no endorsement is inferred from it.

Next transformation

See how Scaling Webinars built a deeper eight-agent operating system around an established delivery model. It shows how a departmental shell can develop once provider routes, knowledge and approval rules are fully specified.

Get your AI blueprint

If sales, content and training all share one overloaded chat, start with operating boundaries. Book a free AI audit to identify the smallest department that has enough context and repetition to prove first.

Source record

• OPENCLAW-DEEP-CLIENT-CASE-STUDIES.pdf, page 14.
• RECAST_CASE_STUDIES_2026_UPDATED.pdf, page 15, document RA-CASES-001, version 2.0.
• RECAST_WEBSITE_MASTER_BLUEPRINT.md, sections 10.4, 11 and 58.