How Dinosaur Vietnam is building an AI operating layer for creative production.
A forward-deployed system for always-on social campaigns, built around brand memory, workflow orchestration, and human review.
Creative Agent
Brand Knowledge, self-hosted n8n workflow orchestration, and chat-based agent interface
1 structured Brand Knowledge layer for brand and campaign memory
4 production workflows inside one agent interface
2 knowledge creation paths: manual entry and AI-assisted extraction
3 core system layers: Brand Knowledge, self-hosted n8n workflow orchestration, and chat-based agent interface
0 generation allowed until required campaign knowledge is confirmed
The Context
A veteran creative agency under monthly production pressure
Dinosaur Vietnam is a veteran creative agency in Vietnam. Its work depends on taste, local nuance, campaign strategy, and the ability to turn a brand idea into social content that clients can actually approve.
For every active campaign, the team has to create a monthly always-on social content plan. The cycle repeats: build next month's content, package it into a client proposal, collect feedback, revise, re-approve, then produce final assets for publishing.
The process sounds simple from the outside. Inside the team, it carries real operational weight.
Each campaign has its own concept, art direction, content pillars, content angles, writing style, do-not rules, approved examples, and rejected examples. Much of this knowledge lives across files, decks, meeting notes, drive folders, and senior team memory.
AI tools had already entered the workflow. But they did not remove the work. They created a second layer of work.
The team still had to craft the creative output by hand. At the same time, they had to train AI tools, rewrite prompts, correct off-brand results, and repeat the same context setup on every project. When an employee left, the prompt knowledge left with them.
Dinosaur did not need another AI generator. They needed a system that could remember the brand, follow the campaign, respect the creative process, and improve through human feedback.
Industry Pain Point
AI is moving fast, but agency workflows are still fragmented
Creative agencies do not sell generic content. They sell judgment. They sell brand fluency. They sell the ability to turn strategy into work that survives client review.
That is why generic AI tools often break down inside agency work.
A social post may look small, but it carries many hidden decisions: which product to feature, which pillar it belongs to, which audience it speaks to, which claim must be avoided, what tone the brand allows, how the visual system should behave, and what the client has already rejected in the past.
When that knowledge is scattered, AI cannot reason with it. It guesses.
The result is familiar across the industry. Junior teams spend hours prompting and reprompting. Art Directors spend time fixing outputs instead of shaping ideas. Supervisors step down into execution because quality is unstable. Clients send feedback that triggers another round of manual revision.
The agency may adopt AI, but the workload does not fall. It shifts into hidden coordination, prompt management, and quality control.
This is the real pain: agencies have brand intelligence, but it is not structured into a system AI can use.
How Lucents Forward-Deployed the Solution
We did not start with an agent demo. We started inside the production workflow.
Most AI vendors begin with a tool. Lucents began with the work.
We sat with Dinosaur's creative team to understand how always-on social production actually moves from campaign input to client-ready output. We mapped the monthly process across both copy and visual creation: product research, content pillars, content angles, writing versions, black-and-white sketches, proposal packaging, client feedback, and final artwork.
That discovery changed the direction of the project.
The first idea was a broad AI OS across six departments. During the forward-deploy process, Lucents and Dinosaur narrowed the scope to the creative team's most painful recurring workflow: always-on social content production.
This was not a downgrade. It was a sharper bet.
Instead of spreading effort across the whole agency, we focused on the workflow that carries daily creative pressure and direct client revenue. The goal was not to automate people out of the process. The goal was to help Dinosaur produce stronger first drafts, reduce repeated setup work, preserve campaign knowledge, and shorten the path from idea to approved asset.
Lucents translated Dinosaur's actual workflow into product logic. Brand knowledge became structured objects. Campaign readiness became a gate before generation. Creative steps became separate workflows. Human feedback became training data for future output.
The result is not a chatbot sitting on top of messy files. It is a creative production system built around how Dinosaur already works.
The Approach
From AI tool to creative production system
Phase 01: Ground-zero discovery
Lucents embedded with Dinosaur's team to map the real monthly content cycle. We studied how a campaign moves from concept to content pillars, from angles to copy, from copy to sketch, and from sketch to final social post.
This phase clarified the main constraint: speed alone is not enough. If the output is not on-brief, on-brand, and usable in a client proposal, the team still has to redo the work.
Phase 02: Brand memory and workflow design
We designed the system around Brand Knowledge, not around prompt boxes.
Each brand and campaign is stored as typed knowledge: brand tone, visual language, color palette, typography, do-not rules, campaign objective, campaign message, target audience, channel requirements, campaign mood, content message house, approved examples, and rejected approaches.
The system supports manual entry and AI-assisted extraction from campaign files. A human confirms, edits, or rejects extracted knowledge before it becomes part of the library.
Phase 03: Agent MVP and workflow orchestration
We designed a single chat-based Agent Interface for the creative team. Users work naturally in chat, then trigger the right workflow with commands such as /workflow/content-angles, /workflow/content-writing, /workflow/sketch, and /workflow/social-post.
Behind the interface, self-hosted n8n orchestrates the generation steps. The Agent assembles the right brand and campaign context, sends it to the workflow, receives the output, and presents it back in the chat for review.
Every approval, rejection, refinement, and regeneration is kept in the session history, creating a feedback trail the team can use to improve future results.
Deliverables
What Lucents built for Creative Agent
Brand Knowledge Function
Dinosaur needed AI that could understand the brand before generating content.
Before Creative Agent, campaign knowledge lived across files, client briefs, internal notes, visual references, and the memory of senior creative people. AI tools could read fragments, but they could not reliably understand the working structure behind a brand or campaign.
Lucents built the Brand Knowledge Function as the system's memory layer. It stores brand-level and campaign-level knowledge as structured objects, with source traceability, version history, and confirmation status.
This gives the creative team a clear answer to a hard question: is this campaign ready for AI generation, or is the required context still missing?
Agent Interface
Dinosaur needed one place for the creative team to work with AI without jumping between tools.
Lucents designed a single chat-based interface where users can describe intent, upload references, trigger workflows, review outputs, approve directions, reject weak results, and request refinements.
The interface does not ask users to become prompt engineers. It lets them work in their own language, while the system handles context assembly and workflow execution behind the scenes.
Content Writing Workflow
Once the angles are selected, the team needs social copy that matches the channel and the campaign.
Before Creative Agent, copywriters created multiple versions per angle, checked tone and restrictions manually, then adjusted based on internal and client feedback.
Lucents built /workflow/content-writing to generate three writing versions per selected angle. Each version includes post copy, hashtag suggestions, CTA suggestions, and channel tagging for Facebook or Instagram.
The workflow pulls from brand tone, campaign message, target audience, writing examples, channel requirements, do-not rules, and previously approved or rejected copy.
Social Post Workflow
After client approval, the team needs final static post visuals that follow the campaign's art direction.
Before Creative Agent, final artwork required designers to reapply patterns, colors, typography, elements, line styles, logo rules, and platform dimensions by hand. Client feedback often focused on details such as character posing, background, lighting, color accuracy, clothing, expression, and framing.
Lucents built /workflow/social-post to generate a final static social post from approved copy and an approved sketch. The workflow applies the full art direction and supports Facebook, Instagram Feed, and Instagram Story formats.
The team can approve, regenerate, resize, or request specific visual refinements inside the same Agent Interface.
Creative Agent Transformation
From repeated AI setup to a reusable creative production layer
Before Creative Agent
Dinosaur's team had to rebuild campaign context again and again. Brand knowledge lived across files, decks, references, and memory. AI tools produced outputs, but the team still had to train, correct, and manage them manually.
Lucents forward deploys
Lucents worked inside the creative process, mapped the real monthly workflow, narrowed the project from a broad AI OS to the highest-pressure social production flow, and turned Dinosaur's production steps into structured product logic.
With Creative Agent
Dinosaur gets a reusable system for always-on social production. Brand and campaign knowledge becomes structured. Workflows become repeatable. Human review becomes data. The team can move from campaign input to angles, copy, sketch, and final post inside one agent interface.
Build AI around the work, not around a prompt box.
Lucents helps agencies and businesses turn messy operational knowledge into software systems that teams can use every day.
Book a discovery call.