​Read online​

Welcome to Unpacking Meaning. If you received this from a friend and enjoy it, ​subscribe here.

The era of human-agent fit

People are talking about how much work they get done with AI agents, but not enough about what you learn from them.

We risk getting into a spiral of productivity-maxing (more oputput), while skipping the fundamentals (more understanding). Don’t get me wrong, I’m all for getting more done, especially with little resources, but what’s different here compared to a lot of tech that drastically changed how we work, is the layers and layers of operational debt you risk accumulating if you do things wrong. When we got spreadsheets, we couldn’t mess it up too much. Still, they completely changed how we went about a lot of our daily tasks. With agents, you’re adding an autonomous operator that learns daily, to your existing systems. The risk of doing things wrong goes up exponentially.

Whether you’re a marketer, founder, or freelancer, I think soon we’ll start talking about “human-agent fit”, or how well you collaborate with your agent.

I’ve been experimenting with my OpenClaw Travis for a month now, and this week I decided to pull the trigger on a second agent, using a slightly different foundation called Hermes Agent. I called him Kobe (Travis Barker and Kobe Bryant are my idols).

As I’m writing this, they’re both on my Discord server in multiple channels, helping me on different areas of my business. It’s still all super clunky and messy, but we’re getting better every day. Most of all, I’m learning a lot about finding my human-agent fit.

Here’s some things I’ve learned in no particular order:

Context quality is the bottleneck.

When I give my agents a vague brief, I get vague output. When I give them the right files, real examples, real voice of customer I collected, they stop rambling with slop and start “thinking” better. Yesterday for example I asked Kobe to go through some strategy docs for a client, and then write a draft of a page as a test. As I was expecting, the output was pretty mediocre. But then when I asked him (yes ask your agents how to work better with them!), he proposed, next time I should share a full example of that page first. I facepalmed. I would have done it with a human, but with this agent I expected him to be this super natural copywriting talent. It’s easy to fall for this.

“Experience” and “expertise” dictate roles.

At first I had Travis doing everything. When I added Kobe, I thought about giving them roles, but then I decided to just watch them work. Their context (memory files) organically shaped each other’s roles without me doing anything. Travis, having more “experience” (time working with me) and “expertise” (knowledge accumulated), naturally became the guide and supervisor. Now he’s the Chief of staff, and Kobe is my CMO/Operator. Here’s a good example of how this dynamic can work well.

Understanding agent architecture helps a lot.

These are not humans (duh), but it’s very easy to humanize them. While that’s fun, learning how agents actually work will give you exponentially better and faster results. That means understanding the basics of their memory, context, reasoning, tasks, and how they work within the tools they use. This is a great guide, but in doubt, just ask your agent. They know themselves pretty well.

Create feedback loops.

The coolest thing I did so far was to replace my daily morning plan and evening review with just chatting to my agents. They store and munch on my thoughts, plans and goals, then mirror them back the day after to keep me accountable and me on track. If they have any thoughts, they also add those to our notes.

Learn to delegate the teaching too.

Sometimes the best results and most productive sessions happen when I let Travis, the more experienced agent, fill Kobe in on projects or processes we have in place. I don’t have to repeat myself and they learn to collaborate better and faster. Bonus: you become a better delegator, especially if you realize the agent you delegate the teaching too, does it wrong. It means you’ve done a bad job initially.

“Hire” a senior engineer.

If you’re a marketer or founder you probably have a Claude account. If not and are thinking about hiring agents, go get one. Why? With a Claude account you also get Claude Code which is the most powerful agent/coder available now. It operates in your terminal or in the native app, and can control your entire computer. This also means it can fix, whatever happens to your agents if they’re running on the same device. Whenever something happens, and your AI team breaks, just type “Claude” in your terminal, and ask it to look into it in plain English.

You’ll need patience.

if you’re thinking of getting into using agents at this stage, expect things to break daily. What do you do when the writing or scheduling task you were relying on doesn’t work? First, have backup plans, and second, be patient, fix it (Claude code to the rescue) and try again. It’s the price to pay for learning this early in the game.

What does this have to do with marketing leadership?

Teaching an AI agent about your business, about your voice, about your buyers, and about your specific way of doing things, is the same skill as teaching a human employee. But then you have to layer on top of it some weirder and often clunkier mechanics. At this stage it might still feel awkward, and unnatural. But in the long run, this will be the future of “leading” marketing teams. When intelligence is the foundation on which your entire company runs, you want to be able to interact with that intelligence, and channel it to interact with other systems effectively.

The time to get your hands dirty is now.

At first, your agent will likely make any delegation skill gaps more apparent and drive you a little nuts. But it will also push you to become a better manager.

Use agents to produce more yes, but also to learn from them.

DISCOVERY

SaaS Half Full 🎙️

In this episode of SaaS Half Full podcast, I break down why we’re entering a “trust recession” in B2B SaaS—and how shallow messaging, driven by surface-level AI use, is making it worse.

We get into what it actually takes to find message-market fit today: raw customer insights, stronger and more opinionated points of view, and using AI as an input machine—not a shortcut for output.

If you’ve ever felt like your messaging sounds right but still doesn’t convert, this one will show you what’s missing—and how to fix it.

How to master brand voice amid the AI slop

“Show me another brand that isn’t friendly, human, and casual.”

You probably can’t.

That’s what Justin Blackman calls “The Trifecta of Nothingness” — the most common brief clients give writers, and the most meaningless. Real voice definition goes deeper than any personality test.

​

Check it out and find the full show notes here.

RESONANCE

Jack Dorsey

Have a great weekend!

Cheers,

Chris

Chris Silvestri Founder & conversion alchemist

🙌🏻 Let’s be friends (unless you’re a stalker)​

When you're ready, here's a few ways I can help

Not sure where to start? Take our free message-market fit scorecard.​