Inside the Intelligence Layer: How Dreamers of Day Operationalizes AI Agents

At WP Engine’s recent DE{CODE} virtual conference, a central theme echoed across every session: the future of the internet is human-first, AI-native. For agency leaders, this shift brings a mix of opportunity and logistical anxiety. AI can proactively audit, optimize, and generate content at a speed that was unimaginable a few years ago.

But this sudden leap in efficiency exposes a hidden operational bottleneck.

“In a world where AI agents gather data, generate content, make recommendations, and execute workflows, the human approval experience becomes increasingly important,” says Mark Atli, Co-Founder and CTO of product and intelligence agency Dreamers of Day.

Speaking at a recent WP Engine Agency Partner Pulse spotlight session, Mark threw down a warning flag for rapidly automating firms: “A single agent can feel impressive, but 10 coordinated agents without strong governance and approval workflows quickly become chaos.”

For more than a decade, Dreamers of Day has managed medium-to-enterprise-scale websites. Over the last year and a half, Mark and his team have quietly architected a proprietary framework to solve the AI scale paradox.

They call it the Intelligence Layer, and it offers a brilliant blueprint for how modern agencies can scale their output while keeping human taste firmly at the center of the product.

The root problem: Fragmented tools and trapped context

Every agency owner knows the frustration of fragmented context. When you attempt to deploy AI workflows or prompt standard models, the output often falls flat because the AI lacks the “big picture.”

According to Mark, an agency’s operational data is traditionally fractured across disconnected software silos:

• The Communication Silo: Crucial design and developmental constraints are buried deep within endless client email threads and Slack channels.
• The Data Silo: Live user metrics and conversion data sit isolated inside Google Analytics 4 (GA4) or Google Search Console.
• The Asset Silo: Design systems, tokens, and standard operating procedures (SOPs) live separately in Figma files and Google Drive documents.

Worse yet, the true competitive edge of an agency—its brand voice, technical judgment, standards, and collective taste—is trapped inside the heads of its senior team members.

“Everyone knows creating standard operating procedures is incredibly important,” Mark notes. “Not only for when new team members join, but also for AI agents to use as context. But it’s rarely documented enough.”

Without centralizing this knowledge and tool-based data, AI agents operate blindly, leading to generic outputs, hallucinations, and disconnected workflows.

Blueprinting the Intelligence Layer

To bridge the gap between siloed tools and human expertise, Dreamers of Day built a unified, three-part system. This infrastructure functions as a centralized ecosystem that sits above individual client builds or standard agency operations.

1. Daily signal sources (The ingestion layer)

The first layer is entirely focused on gathering data. The system utilizes automated ingestion points to pull data daily from live sites, communication channels, and design assets. This includes monitoring user behavior patterns, sweeping client email approvals, tracking competitor positioning, and evaluating developer logs.

2. The knowledge vault (The core)

All ingested signal data funnels directly into the middle of the ecosystem: the knowledge vault. Mark describes this as a dynamic repository that aggregates the total context of an agency or a specific client.

“The knowledge vault gets richer every single day,” Mark says. “The richer it gets, the more valuable it becomes.”

By storing version-controlled, up-to-date documentation on everything from historical code patches to specific animation preferences, the vault transforms disparate knowledge into an operational asset. If a new developer joins the agency tomorrow, they can query the vault and instantly understand technical nuances or choices made on a client’s site two years prior.

3. The specialized agent fleet (The execution layer)

With an accurate, context-rich knowledge vault established, Dreamers of Day next deployed its agent fleet. Rather than relying on a single, generalized AI chatbot, the agency utilizes a fleet of structured, version-controlled scripts designed for specific agency micro-tasks:

• Technical health agents: These workers continuously monitor system performance, parsing server uptimes and WP Engine error logs.
• QA and testing agents: These agents run automated browser scripts—such as Selenium bots—to navigate sites as a human user would, flagging broken layouts or forms.
• Visibility and SEO agents: They scan sitemaps and integrate with tools like Ahrefs or SEMrush to catch missing meta descriptions, broken links, or configuration errors across massive sites.

Taming the ticket avalanche with Overwatch

While an automated agent fleet sounds ideal in theory, executing it presents an immediate operational challenge: an extreme volume of data. When an automated fleet meticulously crawls an enterprise client site with more than 10,000 pages, it can easily generate hundreds of micro-tickets, accessibility warnings, and performance recommendations in a single afternoon.

“The amount of tasks it generates is overwhelming,” Mark explains. “Suddenly, you get into a scenario where you need a human in the loop to triage them all.”

To prevent this digital avalanche from cluttering their clean, human-managed Jira boards, Dreamers of Day developed a custom internal application named Overwatch.

Designed to mimic a visual inbox, Overwatch surfaces every AI-generated alert and client request in an easy-to-read dashboard. Project managers and team leads review the incoming cards daily, acting as the ultimate gatekeepers of quality.

With a single click, a manager can process an alert through three unique governance paths:

1. Accept and auto-assign: For simple, objective fixes (like updating a missing meta tag or correcting a background error log), the manager accepts the ticket and assigns an AI agent to execute the fix automatically within a safe staging environment.
2. Accept with human judgment: If the alert requires nuanced taste or design alignment (such as fixing a broken layout element), the manager moves the ticket to Jira for a human developer or designer to handle.
3. Reject and learn: If the agent flags a false positive, the manager rejects the card. That rejection creates a feedback loop, updating the prompt constraints in the Knowledge Vault so the agent adapts and avoids making the same mistake twice.

“We need to make the human-in-the-loop experience as easy as possible because the amount of things it flags is overwhelming,” says Mark. “We’re even thinking about a future UI where it’s just swiping through cards—yes, no, yes, no.”

Moving staging to production: A real-world win

Does this system actually hold up under the pressure of enterprise client demands? Mark shares a recent project where a client logged over 800 manual content tickets during a major site migration.

By leveraging their intelligence layer alongside the WordPress®¹ Model Context Protocol (MCP), Dreamers of Day put their content optimization agents to work. The agents successfully parsed, updated, and automatically completed 600 of those 800 tickets.

Crucially, none of those fixes went live without oversight. The agents executed the modifications entirely within an isolated staging environment. The AI then flagged the project manager, saying, “Hey, this looks good. Do you want me to push this to production?”

“We had to give the agents a bunch of context, and we went through trial and error telling them ‘don’t do this, do that,'” Mark notes. “But it knocked out a massive amount of work.”

How to build your own Intelligence Layer

For boutique agencies or freelance developers who feel left behind by the rapid pace of AI development, Mark emphasizes that you do not need to build a complex, custom application like Overwatch on day one. Every agency can begin climbing the AI maturity ladder using simple, cost-effective tools.

Step 1: Master the fundamentals

Stop trying to build complex, multi-step agentic workflows from scratch. Spend a few hours thoroughly reading the foundational documentation for developer tools like Claude Projects, Artifacts, and OpenAI GPTs.

Step 2: Build an MVP knowledge vault

Create a local repository or a shared Google Drive folder. Populate it with roughly 10 cleanly structured markdown or text files that detail your agency’s positioning, target audience, precise coding standards, and common design preferences.

Step 3: Route models intelligently

Stop burning your API budget by routing every single micro-task to the most expensive, premium frontier models. Use cheaper, faster models to summarize text and classify basic errors. Reserve high-reasoning, premium models exclusively for complex logic or structural code updates.

Step 4: Establish a simple triage column

Instruct a scheduled workflow to crawl standard client outputs (such as a Yoast XML sitemap or a WP Engine error log) and write the findings directly to a designated “AI Triage” column on a Trello board or a Google Sheet. Have a human manually review and move items into production.

The new frontier of agency value

As AI continues to commoditize basic coding and rote implementation tasks, where does that leave the traditional digital agency?

For Mark, the answer is clear. The value of an agency is shifting away from basic manual execution and moving toward high-level governance, precise architectural curation, and the management of a client’s specific digital context. Agencies aren’t losing their place; they are step-shifting into strategic editors and overseers of highly automated systems.

This evolution doesn’t mean moving away from trusted platforms. Despite the rapid rise of experimental development frameworks, enterprise RFPs still overwhelmingly demand the structural stability of WordPress.

“Enterprise clients are looking for stability in a massive ecosystem that can reliably support their site for the next ten years,” Mark says. “WordPress is absolutely here to stay, but how we interact with it is changing. Personally, I’m logging into the WordPress backend far less than I used to. Instead, we interact with the agents, and the agents safely interact with WordPress.”

Put your agency in the spotlight

Are you looking to grow your agency with a partner invested in your success?

Join the WP Engine Agency Partner Program and win more business, deliver better client experiences, and unlock new opportunities.

1. WP Engine is a proud member and supporter of the community of WordPress® users. The WordPress® trademarks are the intellectual property of the WordPress Foundation, and the Woo® and WooCommerce® trademarks are the intellectual property of WooCommerce, Inc. Uses of the WordPress®, Woo®, and WooCommerce® names in this website are for identification purposes only and do not imply an endorsement by WordPress Foundation or WooCommerce, Inc. WP Engine is not endorsed or owned by, or affiliated with, the WordPress Foundation or WooCommerce, Inc. ↩︎