Key Points
- Orange Logic SVP Misti Vogt argues on MarTech's podcast that content operations belongs to agentic systems, with "brand agents" handling compliance checks before formal review.
- Agents are probabilistic. Compliance and consistency are not, which means handing your brand rules to a model and hoping is a bad trade.
- The real fix is a blend: deterministic software for the rules that must never vary, AI for the parts that need judgment about language, and a human at the decisions that carry consequences.
- Most teams don't have the structured metadata an agent would need anyway, so the agent has nothing to reason about.
The Agent Is the Least Interesting Part
Strip the word "agentic" out of the interview and the argument holds up better. What Misti Vogt is really describing is metadata discipline, and metadata discipline is unglamorous enough that nobody would book a podcast about it.
She's right about the requirement. She's wrong about who should enforce it.
In the season four premiere of Conversations with MarTech, Vogt, SVP of engagement at Orange Logic, makes the case that content operations belongs to agentic systems. Her strongest point is that DAM platforms were treated as storage for finished creative, and that the buyer is shifting back to IT because information architecture and data context now matter more than the file cabinet. Her second point is that every fragment of content needs to be "self-aware": structured data covering why and when it was created, who the audience is, its full lineage of parent and derivative assets, and how it's performing in the market.
What MarTech Reported
MarTech, owned by Semrush, published the episode on September 2, 2026, hosted by Head of Content & Media Mike Pastore. Their take is that marketing teams produce more content than they can manage, and they do it under the EU AI Act, moving copyright law, and the pressure to sound like themselves. Vogt calls the resulting state "permanent triage," which is the most accurate two words in the piece.
Her proposed fix is back-end orchestration. Instead of creatives digging through brand portals and Figma files for the right logo, an organization deploys an agent that knows the brand guidelines, answers queries directly, and runs automated compliance checks on finished files before formal review. The claimed benefit is faster time to market.
Essential Background Reading:
- Content Conductor: AI Content Production for Brands: An overview of how brand knowledge gets structured into a repeatable content pipeline instead of a one-off prompt.
- How It Works: A walkthrough of the Draft, SEO pass, and Editorial pass stages referenced throughout this piece.
Agents Are Bad at Rules
Here's the position we'll defend, and a reasonable person can push back on it: an agent should never be the thing enforcing a rule that has a right answer.
Ask a language model to strip every em dash from a 2,000-word draft. It'll get most of them. Most is the problem. Run the same request tomorrow on the same draft and you'll get a slightly different result, because that's what probabilistic systems do. A regex gets all of them, every time, and it doesn't need to be asked nicely.
Compliance is a category of work where 94% is a failure. Oxford comma or not. Trademark symbol on first reference. Banned phrase list. Heading hierarchy. Schema validity. These have exactly one correct output, and a model that's brilliant at rewriting a flat paragraph is the wrong tool for checking whether the word "leverage" appears anywhere in a document.
We got this order of operations wrong ourselves for a while. We assumed a well-briefed model with a good voice document would apply the mechanical rules as a byproduct of writing well. It didn't. It applied them beautifully in four sections and forgot one in the fifth, and the failure was never in the same place twice, which is a special kind of maddening because you can't build a review habit around a random error.
So the split we now build to is mechanical:
| Layer | What it owns | Why it owns it |
|---|---|---|
| Deterministic software | Banned phrases, punctuation policy, schema validation, heading structure, link integrity, word counts | Rules with one right answer need zero variance, not good odds |
| AI | Drafting, restructuring, semantic keyword work, tone matching, FAQ generation | Language judgment at volume, where "good" beats "identical" |
| Human | Angle, claims, whether the piece is worth publishing, what goes out under your name | Consequence lives here, and consequence isn't delegable |
Knowing what is better done by AI, what is best done by custom software, and what a human needs to review (and making it easy for them to do it at the right time) is key to the future of marketing operations.
Related Content:
- Agency vs. Content Conductor: How agency retainers compare to a structured, brand-aware pipeline on cost and turnaround time.
- Content Conductor vs. Jasper: A comparison of a prompt-based writing tool against a pipeline built on structured brand knowledge.
- Content Conductor vs. Copy.ai: Where a general-purpose AI writing tool differs from a brand-structured content pipeline.
Where We'd Push Back Harder
The compliance-check use case also solves a problem most content teams don't rank in their top five. Nobody's calendar slips because someone grabbed the wrong logo file. Calendars slip because a piece sat in review for eleven days waiting on a director who had four other things due, or because the brief said "thought leadership on procurement" and the writer guessed.
An agent checking finished files is a spellcheck with better marketing. The payoff sits earlier, at the point where the content gets structured in the first place. Compliance at the end catches font violations. Structure at the start determines whether the piece was worth making at all, and no checker downstream can retrofit an intent that was never there.
And the metadata requirement is harder than the interview lets on. "How it is performing in the market" means your analytics have to be joined to your asset library at the individual-piece level. Most teams can tell you how the blog performed last quarter. Very few can tell you which of six derivative assets from one webinar pulled its weight, and fewer still have that written back into the system where an agent could read it.
Reviewers fix sentences and never touch the underlying architecture, because nobody's job description includes "notice that this article has no discernible intent." That's a structural problem, and no amount of editorial attention finds it.
| What Vogt describes | What most teams have | What has to exist first |
|---|---|---|
| Creation intent captured as structured data | A doc title and a due date | A brief template that records angle, intent, and audience knowledge level |
| Defined target audience per asset | "B2B buyers" | Named segments with what they already know |
| Full asset lineage, parents to derivatives | A Drive folder from Q2 | A naming and linking convention someone enforces |
| Market performance attached to the asset | Quarterly channel-level reporting | Per-piece measurement joined to the library |
| Brand guidelines an agent can query | A PDF from the last rebrand | Rules written as decisions, not adjectives |
That last row is where the blend starts paying off. A rule written as "no em dashes" is code. A rule written as "confident but approachable" is a mood board. Only one of those survives contact with a production system.
Next Steps:
- Features: The full list of pipeline, SEO, and AI-optimization capabilities described in this article.
- For Agencies: How the same structural discipline applies across multiple client brands at once.
What to Do This Month
Skip the agent evaluation. Do this instead, in this order:
- Audit your brief template: If it names a keyword and a word count but not an angle, an intent, or what the reader already knows, your metadata problem starts at intake.
- Sort your brand rules into two piles: Mechanical rules go in a checker. Judgment rules go in a prompt. If a rule can't be checked by a script, be honest that a human is still going to have to look at it.
- Write the mechanical pile as rules: "Confident but approachable" is a vibe. Write the RULES: no em dashes, Oxford comma always, never open with a prepositional phrase. That's something software can apply without asking permission.
- Pick one asset family and trace its lineage: One webinar, every derivative. Document what came from what. You'll learn in an afternoon what a mess it is.
- Attach performance to individual pieces: Even a spreadsheet counts. Channel-level reporting cannot tell you which content to make more of.
Skip the lineage step and every AI system you deploy will happily generate the seventh version of an asset you already published, because nothing in your library told it the first six exist. That's already happening at teams running high volume without structure.
Structure, Then the Right Layer for Each Job
Content Conductor™ takes the same position from the opposite end. Brand knowledge gets structured once, during onboarding, then stays active through every stage of the pipeline: Draft, SEO pass, Editorial pass, AI-optimization pass, Schema markup, and CMS staging. The mechanical rules run as code, not as a request to a model that might be having an off day. Voice rules live as explicit instructions rather than adjectives in a PDF. Nothing stages to your CMS without a human gate, because the decisions with consequences attached are the ones we deliberately refuse to automate. The learning loop turns your edits into proposed profile updates, which you approve or skip.
You stay the conductor. AI is the orchestra.
Apply to the founding member program
We're onboarding a small group of founding members at reduced pricing. No credit card. Applying takes about a minute — we reply personally.
