Your voice is a skill — build it like one

Michigan is in the Final Four. I am fired up.

The last time they made it this far, I was there. It was 2018. In San Antonio.

I was traveling with the team then, but not in the film sessions or the huddles. I was in a marketing and social media role but I had my own version of it. Reacting to every game in real time. Adjusting content on the fly. Capturing moments that couldn't be planned.

From that vantage point, you see something most fans never do: the intensity between the games. The focus in the hallways. Staff grinding at hours nobody would believe. Players locked in before the arena lights even come on.

A deep March run isn't one big moment. It's dozens of small ones strung together by people who refuse to let up.

Careers work the same way. The big moments get the attention. The small ones build them.

Speaking of moments and basketball — UConn's game-winner last night was unreal. Go watch it if you haven't.

Most people are still handing AI one writing sample and say "sound like me."

That's not a voice. It's a photocopy.

A real voice skill — the kind that makes AI output actually sound like you across formats — isn't built from one clever prompt. It's built in layers. The Boring Marketer released a walkthrough this week that nails the framework. Here's the architecture:

Layer 1: Gather the right input.
Not one post. Not one homepage. A broader set of material — website copy, writing samples, conversations, interviews. Voice never sits in one line. It shows up across repetition.

Layer 2: Separate brand voice from personal voice.
A founder's tone lives in essays, posts, and interviews. A brand's voice lives in landing pages, ads, product copy, and support flows. Different patterns. Different sources. The builder has to recognize the difference.

Layer 3: Pattern recognition.
Not copying surface style. Looking underneath — how ideas are framed, tone repetition, word choice, sentence rhythm. The things that make your writing yours, even when the topic changes. This is what turns content into a usable voice profile.

Layer 4: Human in the loop.
AI spots recurring signals. The human decides which signals to keep. Without this step, the profile drifts toward an average of everything you've written — including the stuff you'd rather forget.

Layer 5: Package for reusability.
Everything collected, separated, and validated becomes a reusable skill. One that carries voice logic forward so you're not re-explaining the brand every time you need a new asset.

This is exactly what I built with my content skill graph — an 11-node system my agents traverse before writing a single word. Voice rules, platform rules, audience profile, anti-patterns, all linked and layered.

The difference between "AI that sounds generic" and "AI that sounds like you" is the depth of signal you feed it. Most people skip layers 1 through 3 and wonder why everything reads the same.

Your voice is a skill. Build it like one.

The best networking strategy I’ve come to know in this era:

That's it.

Most people network backwards. They ask before they give. They take before they create. They connect before they contribute.

Flip the script. Become interesting first.

We're in a moment where building and shipping in public pays off in ways it never has before. The skills business, this newsletter, the agent systems — all built in public, all generating pipeline I didn't have six months ago.

David Senra shared that Shopify CEO Tobi Lütke told him we're going to look back at 2026 as "the year every single business in the world was up for grabs." AI is coming for everything.

The year it should have been obvious you could rebuild the AI-native version of whatever exists out there.

The builders who share their work publicly are the ones getting found. The ones who build in silence are invisible when the land grab starts.

The best way to build a network is to build something worth talking about. Then talk about it.

OpenClaw vs. Hermes vs. Claude Code.

I've been all-in on OpenClaw for 2+ months. Love the instant text, the scheduled cron jobs for content, research, etc. But Claude Code is still more efficient for a lot of my actual work — and Anthropic keeps shipping features that narrow OpenClaw's advantages.

Now Hermes v0.6.0 dropped from Nous Research. Open-source agent framework. Multi-instance profiles, MCP server integration, provider failover chains, Docker containers, new platform support across Slack, Telegram, and Feishu. 95 pull requests merged in 2 days.

The agent infrastructure space is moving fast. Loyalty to a single platform is a liability. I haven’t done anything yet, but watching this one closely.

$500 Challenge — Final Day.

$474 / $500. 900+ total sales across paid and free. Momentum slowed the past 10 days. One day left to hit my totally random goal. 😃

Whether I hit it or not, the build-in-public experiment was worth every post. Tomorrow you'll know.

Follow: Sahil Lavingia (@shl) dropped The Minimalist Entrepreneur skills — if you're building lean and selling what you know, worth a look.

Watch again: UConn's game-winner — because some moments deserve a second look.

Three simple reminders for today:

Brian