I have logged over 1,000 hours in Claude Code from December 2025 to July 2026. I shipped more than 150 things in Q1 2026 alone, including dozens of skills, agents, CLI tools, pipelines, and automated workflows.
That number gets a reaction. It’s usually one of two things. That’s insane. Or, what did you actually build?
This article is the answer to the second question.
But first a bit about me: I’m Jessica Malnik. I’ve been doing copy and content work for over a decade, including but not limited to website redesigns, landing pages, comparison pages, case studies, cold email sequences, feature pages, and newsletters. I’ve accumulated expertise and judgement over years of in-house and client work, iteration, and watching what converts vs. what sounds good in a doc.
What changed is that I stopped executing that judgment manually every time. I encoded it into a content engineering system. The frameworks, the voice rules, the quality gates, and the research processes are now running as a production environment.
This is what that system looks like including defined inputs, constraints, and outputs.
TL;DR
- I built a 4-layer content engineering system on top of Claude Code: intelligence gathering, production, quality control, and pipeline automation.
- The intelligence layer runs before anything gets written and includes competitor site scraping, product intelligence, SEO analysis, and call mining from sales call transcripts.
- The production layer encodes my actual copy frameworks into 27+ format-specific skills: landing pages, comparison pages, case studies, cold email, newsletter, and LinkedIn just to name a few.
- The QA layer runs a sharpening agent (Cranky Carl) and a content grading rubric on every output before it ships. A record gets appended to the file.
- The pipeline layer chains these steps automatically. Research output feeds directly into writing input. Writing output routes through QA without a human in the middle.
- The judgment behind the system took a decade to accumulate. The encoding sprint took days to weeks.
What a marketing engineering system actually covers
Content engineering is not using AI to write faster. It’s the practice of encoding editorial and strategic judgment into repeatable systems, so the output reflects expert-level thinking. Applied to copy and messaging specifically, content copy engineering covers the full stack: how intelligence gets gathered, how writing gets produced, how quality gets enforced, and how the steps chain together automatically. The result is a production environment with defined inputs, constraints, and outputs.
The difference is real:
- One-off prompting: You paste something into Claude, get a draft, edit it, and ship it. Repeat from scratch every time. Your judgment enters the process during editing. Nothing compounds.
- Content engineering: You build a production environment where your frameworks, voice constraints, and quality standards are encoded in advance. The system produces work that’s 80-90% ready before you touch it. Your judgment gets applied once, at the system level, and scales to every output.
My system covers four layers: intelligence gathering, production, quality control, and pipeline automation. Each one is built from a combination of custom skills, specialized agents, and CLI tools running on my own machine or via Railway apps.
Layer 1: Intelligence gathering before writing a word
The most common failure in content and copy work is starting from a product brief or existing client copy. Both are usually incomplete, internally focused, or wrong in ways nobody has noticed yet.
Before writing anything, I run content intelligence gathering. It covers three areas.
Competitor intelligence
I built a Node.js sitemap scraper that crawls a competitor’s full website and extracts every page title, H1, meta description, and opening paragraph into a structured markdown file. A mid-size SaaS company typically produces 200-800 pages of indexed content in one run.
That file gets piped into a Claude Code skill called scrape-competitor, which analyzes URL patterns to identify content categories and generates a competitive research README with strategic observations.
- What it actually surfaces:
- Which problems they’re positioning around
- What they avoid saying (gaps are as revealing as what’s present)
- How they frame comparisons against competitors
- Which audience signals appear in headline copy
This runs in one command. It used to take half a day.
Product intelligence
For client work, I have an agent called Harriet. She crawls the client’s live site, hunts for a public knowledge base, reads their changelog, and compiles a feature-by-feature product intelligence brief in one pass.
The brief includes product overviews, features found with source attribution, messaging themes, pricing structure, gaps in existing content, and raw copy from every feature page. It’s designed to be the only input a writer needs to produce a feature page without additional research. Anything not confirmed gets flagged with confirm with client before publishing. That quality constraint is built into the research step, not the writing step.
SEO and traffic intelligence
I have a set of Python scripts that cover keyword opportunity analysis, competitor keyword gap analysis, SERP structure analysis, and traffic or content decline prioritization. Each script is built for a specific analytical task and outputs a structured file.
For the SEO refresh agent, it ingests two periods of analytics data from GA4, GSC, or Fathom, categorizes posts by decline severity and traffic volume, and outputs a prioritized refresh queue. It then scrapes and rewrites the top candidates. One command from which posts are losing traffic to a prioritized queue with the refreshed drafts ready for editing.
Call mining and customer intelligence
For client work, the most valuable source of positioning language is almost never the brief. It’s the sales calls and customer interviews.
I process sales call transcripts to extract objections, buying signals, language patterns, and the specific phrases buyers use to describe the problem. That language belongs in the copy, not the polished version of it that a marketer would write, but the actual words a buyer used mid-sentence when they described their frustration.
Customer and case study interviews get the same treatment. The case study skill processes interview transcripts directly, extracting verbatim quotes, identifying the emotional arc, and flagging the moments where the customer’s language is sharp enough to anchor a headline or lead the story. The intelligence layer feeds the production layer. Nothing important gets paraphrased out of existence before a writer touches it.
Layer 2: Production skills as codified judgment
The production layer of a content engineering system is where most people think the work starts. It isn’t. Once the intelligence layer has run, this is where judgment gets encoded into format-specific output.
I have 27+ skills configured in Claude Code. Each one covers a specific content or copy format. Each one encodes my actual process based on proven results.
Landing pages and comparison pages
My landing page skill documents my wedge-to-wow framework in sequence: wedge framing, cost of inaction, solution re-introduction, wow moments, and social proof. That order is non-negotiable because that sequence actually moves buyers.
The comparison page skill goes further. It specifies section order, table row priority, FAQ structure, and a mandatory pre-writing question set. Before drafting a single word, the skill requires confirmation of who the competitor is, what the client wins on, what the competitor genuinely wins on, available customer quotes, and any competitor-specific angles.
- It also runs a self-check before each section:
- Could this copy have been written without knowing who the competitor is?
- Does any sentence resemble the reference file?
- Is the specific competitive dynamic visible in this section?
Fail any check and it rewrites before outputting. That’s judgment encoded as a gate.
Case studies
My case study skill runs in five phases: transcript parsing, narrative arc development, quote strategy, draft writing, and QA. It stops at every phase and waits for explicit go-ahead before continuing.
The first principle it operates from is the customer is always the hero. The product enabled their win. It is never the center of attention. Every case study writer knows this in theory. Most case studies still violate it. The skill enforces it structurally.
Quotes are verbatim or they don’t exist. Cleaned quotes get flagged explicitly. Gaps in the transcript get flagged and a follow-up question gets written. The skill runs my content-grading skill on the final draft before delivering it.
Cold email
The cold email skill won’t write a sequence without proof. If proof is missing, it stops and says so. It has a self-check that runs before delivery. Every subject line must be a specific stat or uncomfortable truth. Every P.S. must do new work. No email can have more than one ask. If any item fails, it fixes or flags it before outputting.
Newsletter and short-form content
These skills have my voice baked in at a level that goes beyond tone instructions. They include specific structural patterns, specific things I don’t do, and specific defaults that reflect how I actually write.
Layer 3: Quality control
The QA layer is what separates a content engineering system from a content generation tool. Skills produce output. Agents evaluate it.
Cranky Carl
Carl is my copywriting rewriter agent configured with specific rewrite rules: generic to specific, vague to concrete, corporate to direct, and convince-everyone to filter-buyers. He has rules and he applies them every time.
This is a mandatory step in the feature page pipeline. The draft routes through Cranky Carl before it gets graded. His job is to make the copy sharp and filtering. He doesn’t soften. He doesn’t negotiate.
Content grading
The content-grading skill applies an A-F rubric before anything ships. An A grade requires:
- Positioning only this person or company could claim
- At least one line a competitor would read and wish they’d said
- Zero sentences that exist just to sound authoritative
Fine but forgettable is a valid grade. It means publishable, not embarrassing, won’t break anything, but also won’t win any awards. The grade gets appended to the output file.
This creates a feedback loop that doesn’t rely on memory. I can look at a file six months later and see what grade it earned and why.
Layer 4: Content production pipeline that chains steps automatically
Individual skills and agents are useful. A chained content production pipeline is where the leverage compounds.
The comparison page pipeline
The process: scrape-competitor generates a competitor research file. The comparison-page skill writes from that file. The draft routes through Cranky Carl Agent. The graded output saves to disk. The comparison page skill explicitly looks for the scraped competitor data file first. Research and writing are connected by a shared file, not by a human copying output between tools.
The feature page pipeline
Harriet Agent runs site intelligence. That output feeds directly into the feature page pipeline. The pipeline writes a full draft, routes through Cranky Carl Agent for sharpening, grades it, and saves one output file with the grade appended.
What makes this work is that the steps are coupled, not just sequential. The research output is designed to be the writing input. The writing output is designed to route through the QA layer. Nothing falls through.
What this is built on top of
This is the part most people skip in articles about AI content engineering systems.
The skills work because the frameworks inside them are real. The landing page skill encodes a structure I’ve used across dozens of clients. The comparison page skill exists because I’ve written enough comparison pages to know exactly what breaks and what converts. The case study skill prevents the LLM from doing what every case study writer does wrong.
None of this works without the judgment behind it. The system is a lever. The fulcrum is the expertise it’s built on.
A well-prompted LLM with no strategic foundation behind it produces polished-sounding nothing. That’s what most AI content looks like. It’s technically correct and strategically empty. It describes products accurately but doesn’t name competitive dynamics or specific buyer friction. It doesn’t convert.
The tooling is not the moat. The encoded judgment is the moat. The tooling makes that judgment consistent and scalable. Those are different things. Copy engineering without the underlying expertise behind it is just fast-moving mediocrity.
What breaks in practice
Skipping the intelligence layer
Starting from a brief instead of evidence produces copy that’s internally coherent but doesn’t connect to actual buyer friction. The landing page sounds fine. It just doesn’t land with buyers. Every hour spent on research saves two hours of revision.
Voice drift
If you stop reviewing outputs and publish everything verbatim, the voice drifts. The system learns from what you publish. If you publish unedited AI output, it starts learning from AI patterns instead of yours. The 10% of editing is not optional. It’s what keeps the system calibrated to your actual voice.
Disconnected research and writing
This is the most common failure I see when people try to build these workflows themselves: the research output lives in one file and the writing step doesn’t reference it, or it references it inconsistently. Every pipeline I’ve built forces the connection explicitly. The steps are coupled by design.
Skills that are too long
Longer skill files are not better skill files. The more constraints you add, the more likely it is that the LLM starts dropping guidance mid-execution. I’ve pruned skills to their minimum effective instruction set multiple times. If a constraint isn’t visibly changing output, it’s noise.
A feedback loop that breaks
If you don’t track what performs, the system can’t improve. Every skill in my stack has either a self-check or a grading step. The SEO refresh agent tracks decline severity. The content-grading skill creates a record. Without data flowing somewhere, you’re generating content into a void.
Checklist: What you actually need to build this
Research layer
- [ ] A way to systematically extract competitor site structure and messaging
- [ ] A product intelligence process that outputs something a writer can actually use
- [ ] SEO and traffic data piped into prioritization logic, not just spreadsheets
Production layer
- [ ] Skills or templates for every format you produce regularly
- [ ] Voice encoded at the level of structure and constraints, not just tone instructions
- [ ] Quality gates built into the writing step, not applied after
Quality control layer
- [ ] A sharpening agent or reviewer with defined rules
- [ ] A grading rubric with recorded outputs
- [ ] A process for what happens when something doesn’t grade well
Pipeline layer
- [ ] Research outputs connected to writing inputs explicitly
- [ ] Writing outputs routed through QA automatically
- [ ] One command per pipeline, not five sequential steps you have to remember
Work with me
I’m a content engineer and conversion copywriter. I work with founders and marketing leaders who need the research done right, the copy written to convert, and the quality enforced before anything ships.
What that looks like in practice: I run competitive and product intelligence before writing anything. I mine sales calls and customer interviews for the positioning language that belongs in the copy. I can produce across every format from landing pages and comparison pages to feature pages, case studies, and cold email sequences with QA built into every step.
You get deliverables, not documents to edit. The system runs on my end.
If that’s what you need: book a call.
FAQs
What is content engineering?
Content engineering is the practice of encoding editorial and strategic judgment into repeatable production systems, so every output reflects expert-level thinking. Applied to copy and marketing, it covers the full stack: how competitive intelligence gets gathered, how writing gets produced against defined constraints, how quality gets enforced before delivery, and how distribution runs automatically. The result is a system where your best judgment scales to every piece. It’s different from content marketing, content operations, and AI writing tools. The distinction is that a content engineering system reflects specific, hard-won expertise. Not a workflow running on generic prompts.
What is content engineering?
Content engineering is the practice of applying software engineering principles, such as modular systems, defined inputs and outputs, quality gates, and automated pipelines to content and copy work. Instead of producing each piece from scratch, you build a production environment where your judgment is encoded in advance. The deliverables come out of the system at a defined quality standard, not from prompting a model fresh each time. It’s different from content marketing, from AI writing tools, and from content operations. The distinction is that the system reflects specific expert judgment, not generic best practices.
How is a content engineering system different from a marketing automation platform or AI content tool?
Marketing automation platforms and AI content tools give you pipes. A content engineering system gives you judgment encoded into pipes. The platform executes whatever you put in the workflows. If your workflows are generic, the output is generic. A marketing engineering system built on real copy expertise, including voice constraints, format-specific quality gates, and research processes connected to writing inputs produces output that reflects that expertise at scale. The system described in this article runs on specific comparison page self-checks, case study rules that enforce structure, and cold email constraints that won’t run without real proof. None of that came from a template library.
Is this just “using AI for content”?
No. Using AI for content means prompting a model and editing the output. Content engineering means building a production environment where the inputs, constraints, quality gates, and outputs are defined in advance. The human role is review and judgment, not construction from scratch.
Do I need to be a developer to build something like this?
For the skills and agents layer, no. Claude Code is accessible to non-developers who can describe their process clearly. The CLI tools require Python and Node.js to work. I built them with AI assistance but had to understand what I was building to spec them correctly.
Why not just use an AI content workflow platform?
AI content platforms give you pipes. They do not give you judgment.
You still have to know what to put in the workflows. You still have to define the quality standards, the voice rules, the competitive framing, the structural constraints. The platform executes whatever you encode. If you don’t have a decade of content and copy experience to encode, you get fast-looking generic output with a clean interface.
This is the gap most teams discover after they’ve paid for the platform and run it for three months. The tool works fine. The output is mediocre.
The skills, agents, and pipelines I’ve built encode specific, hard-won judgment. None of those constraints came from a template library. They came from years of knowing what breaks.
A platform can run a workflow. It cannot tell you what the workflow should be, why it works, or what it’s missing. That’s the gap between a content workflow tool and a content engineering system.
Can this be adapted for client work?
Yes. Some skills are generic and apply across clients. Some are client-specific, encoding that client’s voice, brand constraints, product specifics, and conversion goals. The client-specific ones aren’t transferable. That’s the point. The system produces work that could only come from deep familiarity with that client’s positioning, not from a standard template.
How long did this take to build?
Two different timelines are running here. Clarity Briefs launched in mid-2025 as a content brief tool. The content engineering system described in this article — the skills, agents, CLI tools, and pipelines — was built separately between December 2025 and July 2026. 150+ things shipped in Q1 2026 alone.
The judgment behind all of it took years. The encoding sprint took a few months. Most of the heavy infrastructure is now stable. The right question is not how long does it take to build but what is the compounding value of having built it. The answer to that gets better every week.
What does a content engineer do?
A content engineer builds and operates the production infrastructure behind content and copy output. That includes the research systems, the writing frameworks, the quality gates, and the automation pipelines. The deliverable is not a document or a draft. It’s a system that produces documents and drafts at a defined quality standard. In practice, a content engineer designs the workflow, encodes the judgment, runs the production layer, and reviews the output. The job is building and maintaining the machine, not writing one piece at a time.
Is the content engineer the future of content marketing?
For solo operators and lean marketing teams, yes. A content engineering system can produce what used to require a 3-5 person team. The operators who build these systems absorb work that currently gets distributed across writers, SEO specialists, editors, and strategists. For large teams with headcount, it shifts the job rather than replacing it. The people who can engineer the system become more valuable than the people who just execute inside it. The economics are already moving in this direction. The question is how fast teams are willing to acknowledge it.
How does a writer become a content engineer?
The writing background is the prerequisite. What you need to add is the ability to systematize your own process, enough technical comfort to work with CLI tools and AI systems, and the willingness to think about your work as a production environment rather than a creative output. The fastest path is to start encoding your own judgment. Pick one format you know well, write the constraints and quality gates for it, build a skill or prompt that enforces them, and iterate from there. The system I have took months to build. The judgment it runs on took a decade to accumulate. You need both.
Is a content engineer and a content strategist the same thing?
No. A content strategist defines what to make and why. A content engineer builds the system that makes it and enforces quality. They overlap on research and audience understanding, but a strategist’s primary output is direction and prioritization. A content engineer’s primary output is production infrastructure. Most people in content work are closer to strategists than engineers. A content engineering system requires both which is why it’s hard to hire for separately, and why the people who can do both are in a different category.
How do I hire Jessica Malnik to do content engineering for my team?
Book a call here. Most engagements start with competitive and product intelligence. The system runs on two or three key competitors and the client’s own site before a word gets written. From there, production covers whatever formats you need: landing pages, comparison pages, feature pages, case studies, cold email sequences, etc. QA is built in before anything is delivered. You get deliverables, not documents to edit.
