A recent essay by Matt Schumer racked up over 85 million views on X by making one very uncomfortable comparison: the moment we are in right now with AI is the equivalent of February 2020. Most people had heard something was happening. Almost nobody understood the scale of what was coming next.

It is a provocative framing, and it is fair to push back on the analogy. Comparing anything to COVID is a big swing. But the underlying point is harder to dismiss. The most significant progress in AI right now is not happening where most people are looking.

That is the opening topic Deb Andrews and Brady Lewis tackle in Episode 4 of Candid Conversations: AI & Marketing Unscripted, and it leads into something worth paying close attention to: a working AI agent system, built inside a real marketing agency, that is already changing how client work gets done.

The Progress You Are Not Seeing

If you have been experimenting with AI tools over the past year, you have probably noticed steady improvements. Better writing. Smarter responses. More natural conversation. Those gains are real, but they are not where the transformational shift is happening.

Brady Lewis, Marketri’s Senior Director of AI Innovation, draws an important distinction. The front-end experience most people interact with is improving, yes. But what is changing fastest sits underneath: the infrastructure layer. Models are getting dramatically better at orchestrating tools across multiple applications, spawning subagents to handle tasks in parallel, working autonomously for extended periods, and maintaining memory across complex workflows.

Those are the capabilities that allow someone like Brady to build applications that, in turn, allow non-technical users to do things that would have required a developer two years ago. And because AI can be used to improve itself, the rate of change from here forward does not follow the curve most people expect.

The honest assessment on timing: Brady does not think most people are one month away from a life-changing moment. Technical progress is extraordinary, but the mainstream experience of that progress will arrive more gradually. That said, the gap between people who are building AI fluency and people who are not is widening every week, not closing.

As Deb puts it, working in Claude has become as routine for her as brushing her teeth. That fluency did not happen overnight. It took sustained, daily practice. And for anyone who has not started building that habit yet, the window to catch up is getting smaller.

Project Cockpit: An AI Agent System Built for Real Work

At Marketri, every client account is staffed with a pilot and a co-pilot. The pilot carries more experience. The co-pilot handles much of the day-to-day tactical execution. Project Cockpit takes that model and extends it, building an AI agent architecture designed to handle the bulk of the co-pilot’s routine work.

The system centers on a master co-pilot, an orchestrating agent that serves as the human co-pilot’s primary interface. Beneath it sit a series of specialized subagents, each built to handle a distinct function. The ones currently live include Connie for content creation, Scout for market and competitive intelligence, and Dash for analytics. Six additional agents cover other areas of the co-pilot role.

The human never has to interact with each subagent directly. They work through the master co-pilot, which routes requests to the right specialist and returns a coherent response. Over time, integrations with Google Drive, Sheets, Slack, and PowerPoint will enable fully automated workflows, including generating a monthly report and pushing it to the right folder on a schedule.

The goal is not headcount reduction. Deb is emphatic about that. The point is to free co-pilots from purely tactical execution so they can spend more time thinking, strategizing, and engaging creatively with the work. It is a frustration she has heard across 22 years of agency ownership: people feeling too buried in deliverables to give any single project the strategic attention it deserves.

Project Cockpit addresses that directly. A co-pilot who is no longer manually building out every HubSpot workflow or drafting every content piece from scratch has the bandwidth to write more thorough briefs, explore more creative directions, and think more carefully about what the work actually needs.

Connie and Scout are already live and being used in active client work. The rest of the system, including the master co-pilot and key integrations, will be built out by October 2026, when Marketri plans to present the full journey at Macon.

What This Looks Like in Practice

Two examples from the episode show just how much the gap between idea and finished output has collapsed.

Brady describes a moment most LinkedIn users will recognize. An influencer posts a compelling guide, ten tips on Google Knowledge Panels, but to get it you have to comment, connect, and wait for a DM. By the time it arrives, you have moved on.

His workaround: he went to Claude, pasted the LinkedIn caption, and asked it to research the topic, verify the information was current, and produce a similar guide formatted as a multi-page Word document with a one-page cheat sheet. One prompt. About two minutes. The output was a seven-page guide with professional formatting that Brady says he likely could not have replicated manually in the same time.

Deb’s experiment tackles a different problem. She has been speaking publicly about the shift toward the creator economy, where people increasingly follow individuals rather than companies and websites. If that is where attention is heading, she reasoned, she needed to show up more consistently.

Her commitment was six LinkedIn posts per week, every day but Sunday. As a business owner, that volume felt daunting. Her approach was to go back through her most effective thought leadership writing, upload it to Claude with guidance about her voice and tone, and ask for a series of five posts built around that content. The results matched her voice closely enough to publish as her own. After one month of consistent posting, she gained close to 500 new followers, added a couple hundred newsletter subscribers, and brought in two new leads.

Brady’s observation on why this works matters: Deb did not ask Claude for a generic LinkedIn post. She brought her thinking, her voice, and her strategy. She used AI to amplify and scale what she already had. That distinction is exactly what separates content that resonates from the undifferentiated noise that is already filling every feed.

What This Means for You

A technical tipping point is real, even if the exact timeline is uncertain. The improvements happening at the infrastructure level of AI are moving faster than most people realize, and they are the foundation for everything visible on the front end.

The people not using AI are falling further behind, not staying even. Building fluency takes time and repeated practice. Starting now is not optional.

AI agents can take on tactical execution and free your people for strategic work. Project Cockpit demonstrates that this is not theoretical. It is being built and used right now, at a marketing agency, with off-the-shelf AI tools.

And creativity and voice matter more in this environment, not less. The creator economy rewards distinct points of view. AI can scale and format your thinking, but the thinking still has to be yours.

Watch Episode 4 of Candid Conversations to hear Deb and Brady walk through Project Cockpit’s architecture and demonstrate these use cases in detail.

Watch previous episodes of Candid Conversations or explore our AI and marketing resources for more insights on navigating this transformation.