Amlan Das, ceo @ DAS | audience monetization for retail & ecommerce, recently posted something that made me stop scrolling: "Most teams are still prompting Claude from scratch every session." He added that Claude Cowork Plugins shipped last month and "almost nobody in DTC has explored them yet."

That framing matters. The real unlock is not a slightly better prompt. It is the shift from one-off chat sessions to persistent, specialized roles that carry context forward. In Amlan's words, it is "moving from one-off prompts to an operating system for knowledge work."

Below is my expanded take on what he is pointing to, why it changes the day-to-day for brand and creative ops teams, and how to think about implementing it without getting lost in dense documentation.

The hidden tax of starting from scratch

If your team uses AI regularly, you have probably felt this:

Every session begins with re-explaining the brand, product lines, customer segments, and what "good" looks like.
People save prompts in random places, then copy-paste a "mega-prompt" that becomes outdated in a week.
Outputs vary by who is asking and how much context they remember to include.

That is not just inefficiency. It is operational drift.

Amlan's point is that plugins let you stop rebuilding context and instead "install persistent, specialized roles." Practically, that means you can treat AI less like a search box and more like a team member with a defined job description, inputs, and outputs.

If you want consistent work, you need consistent context.

What an "AI operating system" looks like in practice

When people say "AI agents," it can sound abstract. Here is a concrete way to think about it, inspired by the roles Amlan listed.

An AI operating system for a DTC team typically includes:

Persistent role definitions: A clear purpose, boundaries, tone, and success criteria.
Standard inputs: Brand voice guidelines, product catalog facts, pricing rules, target personas, and current campaign priorities.
Repeatable actions: A set of commands or workflows that produce predictable artifacts (briefs, insights, outlines, QA checklists).
Memory strategy: What should persist, what should refresh, and what must never be stored.
Integration points: Where the output goes next (Notion, Google Docs, Slack, task tracker).

Amlan mentioned his guide includes "full architecture breakdown, slash commands, sub-agents, and the exact code to build 10 specific tools." Even if you never copy his exact implementation, that blueprint is the right mindset: design a system, not a conversation.

Five plugin roles that map cleanly to DTC reality

Amlan highlighted five tools that are especially relevant for brand and creative operations. Here is how I would contextualize each one for a team trying to ship faster with fewer mistakes.

1) Marketing: "Voice DNA" Extractor

The problem: most brands have multiple sub-brands, collections, or audience segments. Without a system, "consistent voice" becomes subjective and outputs swing wildly.

A "Voice DNA" Extractor role should:

Ingest examples of winning emails, ads, landing pages, and social captions.
Extract style rules (cadence, sentence length, banned phrases, preferred framing).
Produce a living style guide that is short enough to use and specific enough to enforce.

The real win is not creating a style guide once. It is making the role the default starting point for every new asset.

Consistency is not a tone prompt. It is a reusable standard.

2) Data: "Dashboard Translator"

The problem: dashboards are built for analysts, but decisions are made by cross-functional teams. Charts do not travel well across the organization.

A "Dashboard Translator" role should turn a screenshot or a set of metrics into:

A narrative summary (what changed, why it matters, what to do next)
A confidence level (what is likely signal vs noise)
Follow-up questions (what data is missing)

This is where persistent context is huge. If the role already knows your North Star metrics, your promo calendar, and your inventory constraints, the narrative gets sharper and more actionable.

3) Product: "Voice of Customer" Engine

The problem: reviews and support tickets are a gold mine, but manually reading thousands of entries is slow, biased, and inconsistent.

A "Voice of Customer" Engine can:

Cluster themes (fit, quality, shipping, packaging, expectations)
Separate "frequency" from "severity" (common annoyance vs deal-breaker)
Pull representative quotes for product pages, ads, and internal reports

The key is to define a taxonomy that matches your business, then keep it stable so trends are comparable month to month.

4) Operations: "Head of Process"

The problem: creative velocity often dies in the in-between steps: approvals, handoffs, unclear ownership, and rework.

A "Head of Process" role should map bottlenecks in approval SOPs, as Amlan put it, and output:

A current-state process map (who does what, when, and with what criteria)
Failure modes (where things get stuck and why)
A revised SOP (fewer steps, clearer definitions of done)
A lightweight QA checklist to prevent avoidable rework

This is the least "sexy" use case, which is exactly why it tends to create the biggest ROI.

5) Strategy: "Board Whisperer"

The problem: leadership meetings create scattered notes. Strategy dies when it is trapped in transcripts and bullet points.

A "Board Whisperer" role can synthesize meeting notes into:

A structured presentation outline
Decisions made, open questions, and owners
Risks and mitigations
A one-page narrative that matches your company’s communication style

Again, persistence matters. If the role knows your recurring board topics, financial cadence, and strategic pillars, it becomes faster and more accurate over time.

Implementation: a practical path that avoids the hype

Amlan said, "There is no shortage of AI hype right now" and called his guide "purely tactical; just copy the code, install it, run it." I agree with the spirit, but I would add one caution: copying tooling is easy; operationalizing it is the work.

Here is a practical rollout plan for a DTC or ecommerce team.

Step 1: Pick one workflow with obvious pain

Choose a task with high repetition and clear outputs, for example:

Weekly performance readout for stakeholders
Review clustering for a hero SKU
Creative brief generation for paid social

If you cannot define "good output" in a paragraph, do not automate it yet.

Step 2: Define the role like a real job

Write a one-page spec:

Mission (what the role exists to do)
Inputs (sources of truth and where they live)
Outputs (format, length, audience)
Guardrails (what it must not do)

This reduces prompt chaos and makes the role transferable across the team.

Step 3: Build command-based usage

Instead of asking the model vaguely, create consistent triggers like:

"/summarize-dashboard" with required fields
"/cluster-reviews" with time range and product
"/draft-brief" with offer, persona, channel

The goal is to make usage predictable so quality is easier to evaluate.

Step 4: Add feedback loops and versioning

Treat the role as a product:

Track examples of great and bad outputs
Update the role definition monthly
Keep a changelog so people trust it (and know what changed)

Step 5: Put governance where it belongs

Some contexts should persist (brand voice, product facts). Others should not (customer PII, confidential financials, legal issues). Decide up front what gets stored and what gets redacted.

Why this matters for creative and brand teams specifically

Marketing teams already have tools. What they often lack is a shared operating layer that turns messy inputs into standardized outputs.

Persistent AI roles help because they:

Reduce rework caused by inconsistent context
Create repeatable artifacts (briefs, insights, outlines)
Make quality less dependent on the individual prompter
Free senior operators to focus on judgment, not formatting

Amlan’s post is a reminder that the competitive edge is shifting. It is not "who uses AI." It is who builds durable systems around it.

Closing thought

If you read Amlan Das’s post and thought, "Cool, but I do not have time to learn another tool," I would reframe it: persistent plugins are less about learning and more about stopping the daily context reset that drains your team.

Once you have one role running well, the rest starts to look less like experimentation and more like infrastructure.

This blog post expands on a viral LinkedIn post by Amlan Das, ceo @ DAS | audience monetization for retail & ecommerce. View the original LinkedIn post →