GlumooPet nutrition

Inside the operating model

We built Glumoo as a brand that learns.

Glumoo is not simply a pet brand that uses AI tools. We connect product knowledge, reusable SOPs, named agent roles, channel workflows and human checkpoints into one operating system—so every completed task can make the next one better.

The short answer

Repeatable work becomes a system. Ambiguous decisions stay human. Agents expand bandwidth; people remain responsible for product facts, brand direction, evidence strength and published outcomes.

The original experiment

From a labour-heavy team to a human + agent operating model.

Glumoo's first AI-native operating experiment documented a smaller core team, higher content capacity and lower generation cost. These are internal operating observations—not customer promises or live counters.

Core team10 → 3people directing agent workflows
Content cadence3 / week → up to 20 / dayacross repeatable content formats
Operating principleSystem firstSOPs and knowledge outlast individual models
Human checkpointAlwaysfor claims, strategy and public commitments

The Glumoo brand OS

A connected operating loop—not a pile of AI tools.

The website is only the public surface. Behind it sits a loop that turns verified inputs into repeatable work, channel output, performance signals and better instructions.

  1. 01Source of truthProduct facts, research, claims, assets and channel rules
  2. 02SOP + Skill layerHuman practice translated into reusable instructions and checks
  3. 03Agent workforceNamed roles with folders, tools, permissions and output locations
  4. 04Channel executionContent, visual production, video, commerce and reporting
  5. 05Review + learningHuman judgment and performance data update the next run

“Models change. A well-designed operating method survives.”

Why start AI-native?

A pet-nutrition brand must manage product specifications, scientific evidence, channel content, customer feedback and operating cadence at the same time. The scarce resource is not more generated copy. It is the ability to connect facts, sources, judgment and publication status.

Glumoo uses agents for work that benefits from standardisation: reading source material, structuring information, comparing versions, checking links, finding conflicts and producing reviewable drafts. This gives people more time for decisions that cannot be delegated.

From source to publication

1. Establish the source of truth

We organise product specifications, source materials, research papers and certification scope first. Uncertain information goes onto a verification list instead of directly into public copy.

2. Organise content by intent

Brand, product, ingredient and decision questions become distinct pages so people, search engines and answer engines can find a clear response.

3. Automate repeatable checks

Agents check titles, descriptions, internal links, structured data, mobile experience and claim consistency, leaving a report that people can review.

4. Review and release

People approve content involving health, product positioning, evidence extrapolation or public commitments before it reaches the live site.

What people must own

  • Define the outcome: decide who the brand serves and which problem it addresses.
  • Judge the evidence: distinguish product facts, research findings, reasonable inference and unsupported speculation.
  • Hold the scope: state whether a certification applies to an ingredient, facility, process or finished product.
  • Own the consequences: remain accountable for publication, product decisions and promises to customers.

How we measure AI-native work

Not by generation speed alone, but by whether information is more traceable, errors surface earlier and each release creates a stronger base for the next one.

What AI must not do

AI cannot invent proof of product efficacy, turn an upstream qualification into a finished-product certification, rewrite correlation as causation or replace veterinary diagnosis. Model output can be wrong, so sources, data and high-risk claims require human verification.

Minimum standard for public content

Do not publish numbers without sources, expand certifications beyond their scope, promise clinical outcomes without direct evidence or release high-risk content without human confirmation.

This approach also shapes the site you are reading: product pages state specifications and boundaries, the NAG guide separates evidence by level, and technical files help search engines and AI systems read the same public facts.

What the system operates

Five workstreams, one shared memory.

Each workflow reads from the same brand facts and returns structured output. That prevents every channel from becoming a separate island.

01 / KNOWLEDGE

Brand knowledge and claim control

Product manuals, research, certification scope, approved language and open questions are organised before content begins. Conflicts go to verification instead of publication.

READ → STRUCTURE → VERIFY
02 / CONTENT

Editorial production for social channels

Topic pools, brand voice, platform templates, review rules and publication records turn “write a post” into a repeatable content operation.

BRIEF → DRAFT → REVIEW → ARCHIVE
03 / VISUAL + VIDEO

Product imagery and the Glumoo Video flywheel

Product assets and creative briefs feed repeatable image and video workflows. Winning hooks, pacing and visual structures are captured for the next round.

EXTRACT → ADAPT → CREATE → LEARN
04 / COMMERCE

Browser-assisted e-commerce operations

Rule-based background work can be automated; exceptions, customer-impacting actions and ambiguous decisions escalate to a person.

OBSERVE → ACT → ESCALATE
05 / INTELLIGENCE

Research, reporting and iteration

Market evidence, channel performance and operating exceptions are turned into reviewable reports that improve priorities, SOPs and agent instructions.

MEASURE → EXPLAIN → UPDATE

Named digital roles

Agents are treated as jobs—not temporary chat windows.

A digital role has a purpose, permitted inputs, tools, a place to put its work, review rules and an escalation path. Names make ownership legible; they do not pretend software is human.

X

Xiaohong

RedNote editorial operations

topics · drafts · publishing records
T

TAT

TikTok video workflow

hook analysis · scripts · video tasks
I

Insley

Instagram visual operations

visual research · asset planning · Reels
E

Agou

E-commerce duty role

browser tasks · checks · exception routing
D

Ashu

Data and reporting role

cross-channel signals · reports · attribution
C

Ama

Engineering and skill role

internal tools · workflow packaging · fixes

Role names and responsibilities document the operating design. The roster evolves as workflows change and should not be read as a real-time staffing dashboard.

Runtime stack

The tools can change. The responsibilities stay clear.

Glumoo separates reasoning, orchestration, persistent execution, engineering and media production. This keeps the operating model portable when a model or vendor changes.

01 · REASONFrontier language modelsResearch, synthesis and difficult reasoning
02 · ORCHESTRATEOpenClawTask routing, schedules and multi-agent coordination
03 · EXECUTEHermes on Mac miniPersistent roles, files, browser sessions and recurring work
04 · ENGINEERClaude Code + CodexInternal tools, website releases, tests and skill packaging

Human × AI protocol

Automation has a boundary. Accountability does not.

Agents can own

Repeatable production

  • Source reading and structured extraction
  • Version comparison and consistency checks
  • Drafting within approved templates
  • Routine browser and file operations
  • Monitoring, reporting and anomaly flags

People must own

Ambiguous judgment

  • Brand and growth strategy
  • Product direction and positioning
  • Evidence interpretation and health claims
  • Relationships and public commitments
  • Permissions, exceptions and final release

The transformation path

Run it once. Turn it into a method. Let the method improve.

  1. 01

    Choose one real workflow

    Start with a repeated bottleneck whose inputs, outputs and owner are visible.

  2. 02

    Establish facts and boundaries

    Collect the source material, approved language, permissions and red lines.

  3. 03

    Run with human checkpoints

    Use real work, record failures and keep review close to high-risk decisions.

  4. 04

    Distil the SOP into skills

    Convert what worked into reusable instructions, tests and output contracts.

  5. 05

    Feed results back

    Use performance and exceptions to improve the knowledge base and next run.

Plain-language answers

Questions about Glumoo's AI-native model

What does “AI-native” mean at Glumoo?

It means designing work around shared knowledge, reusable workflows, agent roles, tests and human checkpoints from the beginning—not adding a chatbot after the process is already built.

Is the company fully autonomous?

No. Agents can execute repeatable work, but people set goals, approve permissions, judge evidence, handle exceptions and own every public outcome.

What is a digital employee?

It is our operating shorthand for a persistent agent role with a defined job, tools, files, permissions, outputs and review rules. It is software under human management.

Can AI approve product or health claims?

No. AI can locate sources and flag inconsistencies, but health-related claims, certification scope and evidence extrapolation require human verification.

How does the system improve?

Completed work, performance signals, corrections and exceptions are recorded. People decide which lessons should update the knowledge base, SOP or agent instructions.

See the method applied

Start with one formula and inspect how facts, evidence and boundaries stay aligned.

View a product page Read the NAG evidence