Build 07 / Planned Agentic Content Engine
Weekly Build Agent
A planned human-gated agent system that turns Tanya Build Cockpit memory and Greg's weekly build sessions into approved content outputs: a build summary, pl8ypus blog draft, Brevo email draft, LinkedIn draft, and approval checklist.
Greg's take
The problem is not creating more content. The problem is turning real build work into public evidence without drifting into fluff, leaking private details, or making claims the work cannot support.
That is why Tanya memory matters. The Weekly Build Agent should not write from vibes or loose chat history. It should write from the build record: what changed, what shipped, what proof exists, what stays private, and what deserves to be said publicly.
This is the content engine I actually want - not an autonomous posting bot, but a controlled evidence-to-content system with a hard human gate.
The problem
Build work creates proof. Most of that proof disappears.
Evidence gets lost
Build reports, screenshots, decisions, fixes, and trade-offs stay scattered across chats, repos, notes, and memory.
Content becomes generic
Without a structured evidence layer, weekly marketing output turns into vague updates instead of credible proof of progress.
AI can overclaim
A writing agent without a safety and proof gate can make private, exaggerated, or unsupported statements sound polished.
Publishing needs control
The system should prepare drafts and evidence packs, but Greg must decide what is safe, useful, and ready to publish.
What it is
A weekly memory-to-content operating layer.
The Tanya Build Cockpit records what happened. The Weekly Build Agent turns that record into a structured weekly pack. The human approval gate decides what can move downstream.
Memory capture
Uses Tanya Build Cockpit as the structured source for weekly build records, decisions, proofs, screenshots, and next steps.
Agent orchestration
Routes the same evidence through specialist agents for intelligence, planning, drafting, and safety review.
Content pack generation
Creates a weekly summary, blog draft, Brevo email draft, LinkedIn draft, and approval checklist from one shared evidence base.
Safety and proof gate
Checks for overclaims, private client details, secrets, unsupported statements, and anything that should not be published.
Human approval
Greg reviews, edits, approves, rejects, or regenerates the weekly pack before anything leaves the system.
Publisher layer later
Future publisher agents can create drafts in site, Brevo, or LinkedIn workflows, but only after explicit approval.
Agentic architecture
Tanya memory in. Approved weekly content out.
The diagram shows the planned system spine: Tanya Build Cockpit memory, Greg's build sessions, evidence collection, specialist agents, safety review, human approval, and controlled downstream outputs.
Click to enlarge
Uses the dark-mode system diagram at images/007-agents.png.
Agent layer
Specialist agents. One shared memory source.
01
Source Collector Agent
Gathers weekly evidence from Tanya memory, build sessions, screenshots, files, commits, and page updates.
02
Weekly Intelligence Agent
Summarises what changed, why it matters, what proof exists, and what should be kept private.
03
Content Planning Agent
Selects the weekly content angle and decides how the evidence should be shaped for each channel.
04
Blog Writer Agent
Drafts a root-level pl8ypus build article with proof, narrative, schema-ready structure, and Greg's take.
05
Email Writer Agent
Creates a Brevo-ready email draft with subject line, preview text, body copy, CTA, and segment notes.
06
LinkedIn Writer Agent
Produces a punchier public post from the same weekly evidence, without turning the build into empty thought leadership.
07
Safety / Proof / Privacy Agent
Flags overclaims, secrets, client-sensitive details, unsupported claims, and public-demo risk.
08
Publisher Agents - later
Creates or publishes approved outputs only after review. The default first version creates drafts, not live posts.
Human in the loop
The human approval gate is not a nice-to-have. It is the control system.
Approval rule
Nothing publishes automatically.
The system can draft, summarise, score, and stage content. It cannot decide what is safe, true, strategic, or ready. Greg owns the final decision.
Approve
Move the weekly pack to draft creation or publishing workflow.
Edit
Adjust claims, tone, framing, CTA, privacy boundaries, or channel format.
Reject or regenerate
Reject weak drafts, request a stronger angle, or hold the story until the build evidence is stronger.
Approved outputs
One evidence pack. Multiple controlled outputs.
Weekly build summary
A concise record of what changed, what shipped, what evidence exists, and what the next build priority is.
pl8ypus blog draft
A root-level build article using proof-led storytelling, SEO structure, and a clear Greg-style point of view.
Brevo email draft
Subject line, preview text, email body, CTA, and segmentation notes prepared for review, not automatic sending.
LinkedIn draft
A sharper social version of the same evidence, kept human, direct, and grounded in the actual build.
Approval checklist
A final control check for claims, privacy, client references, public-demo status, CTA, and publish readiness.
Roadmap
Start with memory and drafts. Earn automation later.
Manual weekly memory pack
Greg records the week's build evidence inside Tanya Build Cockpit. The agent generates drafts from that memory.
GitHub and evidence retrieval
The system pulls stronger evidence from commits, changed files, build reports, screenshots, and approved page updates.
Approved draft creation
After human approval, the publisher layer creates a blog draft, Brevo draft, or LinkedIn-ready copy pack.
Operational content engine
A stable weekly rhythm where real build work becomes public-safe content without becoming automated noise.
Greg's take
A content engine without memory is just a copy machine. It can write words, but it cannot prove anything.
The interesting part of this build is the spine: Tanya records the work, the agents turn it into a weekly pack, the proof agent checks the claims, and the human decides what survives contact with the public internet.
That is the difference between posting more and building reputation. One is volume. The other is evidence.
The cockpit records the work. The agent shapes the story. The human protects the trust.
Want to discuss the Weekly Build Agent?
For architecture conversations, human-gated content agents, build memory systems, or evidence-led AI marketing workflows.