Koka Sexton

TL;DR

• 12 automated cron jobs run my content operations, CRM pipeline, and quality control across 5 content properties without manual intervention
• The system catches encoding corruption, broken links, banned words, and underperforming content that human review would miss
• Each job runs in an isolated agent with defined permissions — some fix silently, others report only
• The key breakthrough was a single instruction change that took three days to figure out: agents default to describing work instead of doing it
• Real result: a banned word in a 3-month-old article caught and fixed automatically, three weeks before I would have noticed

---

Most marketers wake up and check dashboards. I wake up and check what the system did while I was asleep.

That’s not a flex. It’s a design choice.

Six months ago, I spent mornings on the same things every marketer does: scanning analytics, checking for broken links, reviewing content quality, updating CRM records. The routine wasn’t hard. It was just repetitive. And repetition is exactly what automation was built for.

So I built 12 cron jobs that now handle all of it.

These aren’t simple “post at 9 AM” schedulers. They’re decision-making agents that scan content, detect problems, fix what they can, and report what they can’t. They run on weekly cadences across five content properties. They’ve caught encoding corruption I never would have found manually. They’ve flagged problems in articles I’d completely forgotten existed.

And they do it while I sleep.

This is the architecture behind automation that actually works — not the kind that sounds good in a pitch deck, but the kind that catches real problems in real content, week after week, without anyone touching a button.

The Quality Layer: 4 Weekly Content Sweeps

My content sits across five properties. Each one has different formatting rules, different banned word lists, different component requirements. Manually checking every site’s three most recent posts takes about 90 minutes per week. I’ve automated it to zero.

Monday: KSB Sweep. The main site gets the first check. The agent pulls the three newest posts and runs them through four filters: encoding (is every apostrophe an ’ and every em dash an —?), banned words (a growing list of terms I never want in my writing), broken links (every external URL validated), and component completeness (stat cards, TL;DR sections, and CTAs all present).

It caught something three weeks ago that still bothers me. An article from March used a word I’d banned from my writing — in a subheading I’d never noticed during original editing. The sweep flagged it, fixed it, logged the change. I only found out from the weekly summary report. Three months that word sat there in a published article, and I had no idea.

The quality sweeps fix the kind of mistakes you can’t see anymore because you’ve read your own writing too many times.

Tuesday: CCM Sweep. Chief Content Marketer has different rules than KSB. Different banned words. Different component requirements. The Tuesday sweep checks the three most recent CCM posts against the CCM rule set. Encoding issues are the most common catch here — CCM articles pull in external sources more often, and those sources frequently carry non-standard characters that need HTML entity conversion.

Wednesday: MWC Sweep. Mayor of Walnut Creek operates on a different format entirely. Local news structure, different post layout, different quality standards. The Wednesday agent knows the MWC template and validates against it.

Thursday: VCO Sweep. Visibility Creates Opportunity rounds out the week. Same logic, different rule set. By Thursday evening, I’ve had four content properties audited in four days. Each sweep takes roughly 90 seconds of actual processing time. The summary reports hit my notification channel and I skim them in 30 seconds flat.

That’s 90 minutes of manual work eliminated — every single week.

The Traffic Layer: 3 Weekly Optimizers

Quality sweeps prevent decay. Traffic optimizers actively improve what’s already published.

Each Monday, Wednesday, and Friday, a different agent pulls analytics data for one of the three main content properties. It finds posts that are getting impressions but underperforming on click-through rate compared to the site average. Then it generates suggestions: title optimization, meta description rewrite, content refresh triggers.

The Monday KSB optimizer looks at the last 90 days of search data, identifies the three worst-performing posts by CTR relative to expected CTR for their position, and generates specific revision suggestions. Sometimes it’s a title that needs more urgency. Sometimes it’s a meta description that reads fine on paper but doesn’t earn the click.

The Wednesday CCM optimizer does the same for chiefcontentmarketer.com. The Friday MWC optimizer handles mayorofwalnutcreek.com.

These three agents don’t auto-publish changes. They report recommendations. The risk-reward ratio on auto-modifying published content is too high — one bad title change on a ranking page could undo months of SEO work. Report-only, by design.

Here’s what matters: I never have to go looking for optimization opportunities. They come to me, prioritized, with data attached. I spend five minutes reviewing suggestions instead of 30 minutes hunting for underperformers.

The Pipeline Layer: CRM Sync and Health Monitoring

Content and traffic sit at the surface. The pipeline is what generates revenue. These two jobs keep the demand generation engine running clean.

Friday CRM Sync. Every Friday evening, my CRM gets updated. New leads from the week get enriched with company data, contact details, and lead scoring. Duplicates get flagged. Stale records from the prior week get a nudge for follow-up. This used to be a 45-minute Friday afternoon ritual of cross-referencing spreadsheets. Now it happens while I’m eating dinner.

CRM Health Monitor. A separate job runs a data integrity check. Are there contacts with missing email addresses? Companies with incomplete enrichment data? Leads sitting in “new” status without a follow-up task for more than 5 days? The report lands every Friday with a simple color system: green means clean, yellow means review, red means something broke and needs immediate attention.

I’ve had exactly one red report in two months. An automation scenario had updated incorrectly and stopped enriching new leads. The health monitor caught it within five days. Without it, I might have gone weeks before noticing the data gap — and weeks of unenriched leads sitting in the pipeline.

The Content Layer: Research and Asset Management

MWC Research Agent. Runs mid-week, sources trending local stories and discussion threads, compiles a ranked list of topic ideas with source links. It doesn’t write anything. It feeds the idea pipeline so that when I sit down to write, I’m choosing from options rather than staring at a blank page.

Image Auto-Rename. This one sounds small but compounds. Every image that lands in my content drive gets automatically renamed according to a consistent convention. No more “IMG_4728.png” in my media library. Every file follows the same pattern before it ever touches WordPress. Saves maybe 15 seconds per image. Times hundreds of images per year. The math works.

The Self-Improvement Layer

The twelfth cron job doesn’t touch content, traffic, or pipeline. It audits the system itself.

Every Sunday night, a meta-agent reviews the past week’s operations. How many sessions ran? Total cost? Which agents produced errors? Which tasks took longer than expected? Are there recurring failure patterns?

This job produces a one-page report that tells me whether the system is getting better or worse over time. Cost per task. Error rate. Completion rate. Mean time to resolution for flagged issues.

Last month it surfaced a pattern I would have never caught manually: three out of four weekly sweeps were detecting the same encoding error — em dashes not properly converted to HTML entities in a specific content source. I fixed the source template once, and all four sweeps stopped flagging it. Without the meta-analysis, I would have kept catching and fixing the same symptom on repeat, indefinitely.

---

The Architecture That Makes This Work

Each cron job runs as an isolated agent. No shared state. No cross-contamination. If the Tuesday CCM sweep fails, Monday’s KSB sweep isn’t affected. This isolation was intentional — it means one broken job doesn’t cascade into a broken system.

Each agent has defined permissions. Some have read-write access to content databases. Some are read-only by design. The traffic optimizers, as I mentioned, are report-only — the fix involves editorial judgment, not mechanical replacement. The quality sweeps have write access because the fixes are deterministic: replace this character with that character, remove this banned word, fix this broken link. There’s no ambiguity.

This permission model matters more than you’d think. It prevents automation from making judgment calls on things that require judgment, while letting it handle the mechanical work it can do perfectly.

The Bug That Cost Me Three Days

When I first built these agents, they worked perfectly in testing. They would analyze content, generate fix suggestions, produce beautiful reports. Then I’d go check the actual content database and nothing had changed.

The agents were operating in a mode where they described what they should do instead of actually doing it. They’d write paragraphs about the encoding fix they should apply, the banned word they should remove. But they never touched the data.

It took three days to figure out what was happening. The fix was a single instruction: use tools, produce output, don’t describe what you’d do — just do it. Without that explicit directive, the agents defaulted to analysis mode. With it, they started producing actual changes in the real world.

This is the kind of thing you only learn by building. No documentation warns you that your automated agents might talk about the work instead of executing it. The output looks correct — the reports read perfectly — but nothing changes in production. It’s a subtle failure mode that’s easy to miss because everything appears to be working.

The Compounding Effect

Each cron job’s output feeds other systems.

The quality sweeps don’t just fix content. The fixes get logged, and the logs feed the self-improvement analysis. The traffic optimizers don’t just suggest changes. The approved changes get tracked, and the before-and-after CTR data feeds back into the optimization algorithm. The CRM sync doesn’t just update records. The enrichment data feeds into lead scoring, which feeds into outreach prioritization.

Twelve independent jobs, but their outputs connect. A quality fix on a KSB article that improves readability might indirectly improve CTR, which the Monday optimizer notices three weeks later, which generates a title optimization suggestion, which gets approved, which further improves performance. The system compounds across weeks, across properties, across layers.

That’s the real power. Not any single cron job. The fact that once they’re running, they don’t stop. Week after week, they catch things you’d miss, fix things you’d delay, and surface problems before they become emergencies.

What This Means For You

You don’t need 12 cron jobs. You probably don’t need 5.

What you need is to identify the highest-repetition, lowest-judgment tasks in your marketing workflow and automate those first. Start with one: a content quality check that runs once a week, a CRM sync on Fridays, a broken link checker. Pick the thing you dread doing manually. Ship it. Watch it run. Then build the next one.

The barrier isn’t technical. It’s the mental shift from “I should check that” to “the system checks that.” Once you make that shift, the hours start compounding in your favor.

Automate the routine. Save your brain for the decisions.

More on marketing systems: Automation with Intent • Demand Generation