The Build I’ve Been Waiting to Start

Hello and welcome! I’m fresh off a very jetlagged return from Greece, but excited to keep learning out loud and get back to it. This week’s edition is a little different—I’m wrapping up the DeepLearning.AI course on Agentic AI and sharing the key insights I took away from completing it. And now that the course is behind us, I’m finally ready to start building an agentic workflow of my own that you can replicate (more on that below).

Summary: Nearly 80% of B2B brands now use AI for content writing and 70% for creative ideation, positioning it as a collaborator rather than a replacement.
Why it matters: Treating AI as a “sparring partner” keeps the human insight that fuels creativity — a mindset shift that separates thoughtful adopters from lazy automation.

Summary: Many B2B marketers say data silos and poor integration are blocking them from realizing AI’s potential.
Why it matters: Without clean, connected data, even the best AI tools are useless — the smartest marketers will fix their data pipelines before chasing shiny new models.

Summary: AI is helping marketers shift focus from short-term lead gen to long-term customer lifetime value, tying creative work directly to revenue.
Why it matters: Measuring what really matters — retention and ROI — is how AI turns marketing from a cost center into a business growth engine.

We’ve reached the final module in the Agentic AI Deep-Learning course. Over the past few weeks, we’ve broken down how agents think, plan, and execute work. This last module shows how those pieces come together to create systems that can operate with real autonomy, not just handle single tasks.

Instead of asking a model to do everything in one shot, you have it outline the steps first, then execute each step with the right context.

It’s a simple pattern that produces more predictable, higher-quality results—almost like giving the model its own project plan.

Action: Split planning and execution. Use one prompt to map the steps, another to run them.

For more complex work, have the model generate its plan in code instead of plain text.
Code forces structure, reduces randomness, and consistently outperforms free-form planning.

Where this helps: research pipelines, multi-stage content workflows, or anything that needs consistent repetition. Proven to increase the quality of AI outputs when compared against plain text.

Instead of one agent doing everything, you create a small “team” of agents—each responsible for a specific slice of work.

A manager agent coordinates research, writing, QA, and review agents. They pass state, check each other’s work, and mirror how real teams operate.

Use case: A PMM agent that kicks off research, drafts messaging, routes it for refinement, then runs a final quality pass.

A. Linear

B. Hierarchy

C. Deeper Hierarchy

D. All-to-All

Next week, we’re starting the Experiment Log series (previously Project Playground) with a new build: Signal Scanner—an agentic workflow marketers can use to surface the most important industry news and thought-leader insights without the manual digging.

If your role depends on staying ahead of trends or spotting signals early, this is a system you’ll be able to plug straight into your own workflow.

Agentic Loop

The cycle an AI agent follows to get work done: it plans, acts, checks what happened, and adjusts before taking the next step.

This loop is what gives agents autonomy—it lets them handle multi-step tasks without you micromanaging every move.

As always, stay curious and have fun!

Best,
Skyler Neal