Proposed workflow · 14 steps

How the Go Live pipeline works

One recording goes in. A month of content comes out. Nine specialists do the work, coordinated by a tenth agent; five checkpoints belong to you. Nothing reaches an audience until you say yes.

14

Steps end to end

5

Points where you decide

34+

Assets per recording

0

Things that publish without you

From recording to published content, with five human checkpoints Ingest the recording, approve chapters, extract hooks, then approve hooks. Work splits into parallel branches: rank clip candidates, choose ten, and render shorts; draft posts, blogs, and graphics. Review the finished batch, prepare the schedule, and approve release before publishing. Monitor published content for seven days and send lead alerts. Revisions return to the relevant worker; holding release keeps content unpublished. 01 · PREPARE THE SOURCE Ingest recording Transcript + chapters 1. Chapters OK? You review the structure Extract ideas Hooks, quotes + topics 2. Hooks OK? You approve the direction Revise chapters Revise hooks 02 · CREATE IN PARALLEL Rank clip candidates ~25 moments + rationale 3. Pick ten clips You choose the moments Render shorts Crop + captions Re-score with your notes Draft posts, blogs and graphics Three workers run alongside the clip branch; they do not wait for clip selection. Both branches ready → review finished content 03 · REVIEW 4. Approve the batch You review the finished assets Send back with notes to the relevant worker 04 · SCHEDULE, THEN RELEASE Prepare schedule Channels + dates 5. Release? Your final yes Publish Approved schedule Engage 7 days → lead alerts Hold / revise schedule: nothing publishes
Agent does the work
You decide
Revisions return to the relevant worker
Nothing auto-approves.

01 — Agent

Automated

Ingestion Agent

Picks up the recording from SharePoint. Produces a full transcript with word-level timestamps, chapter breaks, and speaker labels.

In

One-hour video in SharePoint

Out

Timestamped transcript, chapters, speakers

The transcript keeps every moment traceable. Speaker labels identify who said what; timestamps link the text back to the recording. Processing time depends on the recording.

Human gate

You decide

Confirm the chapters

The pipeline shows you the chapter breaks. You adjust if needed. If you send it back, Ingest re-runs with your notes. Review time depends on how much needs changing.

Ingest
Chapters?
Extract

This is the pattern. Work flows until it hits a decision. Approve and it continues. Send back and the relevant agent revises its work.

02 — Agent

Automated

Extraction Agent

Reads the transcript and pulls out hooks, quotes, topics, audience questions, and emotional tone.

In

Timestamped transcript with chapters

Out

Hooks, quotes, topics, questions, sentiment

Every idea stays tied to its source. Each hook includes a timestamp; each quote includes its speaker.

Human gate

You decide

Pick the best hooks

You see the top candidates. Pick what fits this client's audience. Say no and Extract re-runs with your feedback.

Extract
Hooks?
Repurpose

Your hooks set the direction. Once approved, clip curation and the posts, blogs, and graphics workers can begin in parallel.

03 — Agent

Automated — owned judgement

Curator Agent

Scores candidate moments on six visible dimensions and ranks them with a rationale. The initial plan targets about 25 candidates from a specialist video service.

Self-containment

Zero context needed?

Hook strength

First 3 seconds grab?

Payoff

Resolves or stops?

Buyer relevance

Right person hears it?

Boundary clean

Sentence-aligned?

Delivery energy

Alive or flat?

The rubric is visible so you can correct the reasoning. Vendor support for timestamped candidates still needs validation on a real recording. Saved selections and notes can inform later scoring. Improvement must be checked against the editor’s judgement.

Human gate

You decide

Pick 10 clips from ranked list

About 25 candidates, scored and explained in a review page. You choose 10. If you're not happy, Curate re-scores and you pick again.

25

Ranked candidates

10

You curate

Saved

Selections + feedback

Your choices teach us what fits this client. Choosing #14 over #3, with a note explaining why, gives more useful feedback than a simple approval.

04 — Agent

Automated — mechanical

Shorts Agent

Takes your 10 moments. Crops to vertical. Tracks the speaker. Burns in captions.

In

Your 10 chosen moments

Out

10 vertical clips with captions

Rendering starts with a specialist service. It uses your selected moments; posts, blogs, and graphics can progress while you make that selection.

05 — Agent

Automated

Posts Agent

Hooks and quotes become 20+ posts, each shaped for its platform.

In

Hooks, quotes, topics

Out

20+ posts total, adapted across platforms

Starts after hook approval. Posts, blogs, and graphics run alongside the clip branch. The batch review waits for the finished assets.

06 — Agent

Automated

Blogs Agent

Groups approved topics into 4 articles. Each develops a clear theme from the conversation and gets a writing and editing pass.

In

Topics, quotes, transcript

Out

4 blog articles

07 — Agent

Automated

Graphics Agent

Best lines laid out as branded images, sized per platform.

In

Approved quotes, client fonts, colours + logo

Out

Quote graphics

Human gate

You decide

Approve the batch

A card in Teams. Approve all, review each, or send back with notes — the relevant agent redrafts and the cycle repeats.

Repurpose
Approve?
Schedule

This approves the content, not its release. After revisions are accepted, the next step prepares the schedule. A separate final approval authorizes publishing.

08 — Agent

Automated — human-gated

Distribution Agent — prepare the schedule

Prepares the destinations and publishing times for approved content. Nothing is sent to LinkedIn, YouTube or X until you approve release.

In

Approved clips, posts, blogs, graphics

Out

Proposed schedule awaiting release approval

The same agent publishes after release approval. It records each published item and checks for duplicates when retrying a delivery.

Human gate

You decide

Release it

The final yes. The card shows what's queued, where it goes, and when. If you hold, nothing publishes.

Schedule
Release?
Publish

09 — Agent

Automated

Engagement Agent

Watches comments for 7 days. Scores responders for buying signals. Alerts you when someone is worth a conversation.

In

Published posts with platform IDs

Out

Lead alerts, engagement metrics

LinkedIn is assisted, not automated. The agent prepares a queue; a person works through it.

The full picture

Nine specialists, one coordinator, five human checkpoints.

The coordinator tracks progress, starts the next workers, and waits for your decisions. Revisions return to the relevant worker. Holding release keeps content unpublished.

10

Short vertical videos

20+

Social posts

4

Blog articles

Planning estimate: ~$125/month for the first client

Shared hosting, database, queue + storage$81/month
Transcription, writing, rendering + delivery$10/recording
Candidate scoring allowance$1/recording
Four recordings: $81 + (4 × $11)$125/month

USD planning assumptions from the design and curator plan, not a vendor quote. Shared infrastructure is not charged again for every client. Excludes development and human review; validate service pricing and candidate availability before committing.

1 / 16