Discovery is dying, AI agents are breaking in weird places, and a teenage intern is quietly outbuilding your growth team: your playbook needs a rewrite.

This week's edition covers:

  • Kevin's Take: the AI sameness problem is real, but cranking up the weirdness dial won't save your brand on its own
  • The Signal: Signal: fresh G2 data on how buyers actually discover B2B software now, plus what 63 million tool calls reveal about where AI agents quietly break
  • Tools & Tactics: how Anthropic's growth team runs build-vs-buy decisions, and the paid-UGC engine behind a $2B AI startup
  • Quick Links: GEO vs SEO across 160 sites, the rise of the GTM engineer, Nielsen on AI ad mentions surging while trust lags

Kevin's Take

The AI sameness problem is real, but weirdness isn't the fix on its own

I think Sam Jacobs is onto something important in this piece on flesh and blood marketing, which is that as everyone pipes their content through the same four or five models with roughly the same prompts, the output converges and distinctiveness becomes the scarce thing. I agree with the diagnosis. Where I'd push back a little is on the fix, because I think a lot of CMOs are going to read weirdness as a license to turn up the brand voice knob and call it a day, and that's not actually what's going on here.

The reason everything sounds the same isn't that the models are boring, it's that most of us are using them the same way, same base model, same vanilla prompts, same lack of any real context about the company, the customer, or the point of view the brand actually holds. When we started building skills inside Claude for our own work at Mighty and True, the thing that changed the output wasn't asking it to be more creative, it was loading it up with real material, our tone of voice docs, actual customer transcripts, the specific way we talk about a problem internally, references to work we admire with notes on why we admire it. The model stops reaching for the median internet answer because you've given it a narrower, weirder, more specific place to reach from.

Weirdness as an output is really just specificity as an input, and most teams haven't done that work yet.

So if you're reading Sam's piece and nodding along, the question I'd sit with this week isn't how do we get weirder, it's what does our AI stack actually know about us that a competitor's stack doesn't know about them.

If the answer is nothing, we're going to sound like everyone else no matter how many times we tell the model to be bold, because we're all drawing from the same pool.

The distinctiveness has to exist somewhere before the model can express it, and right now for most B2B companies it only lives in the heads of three or four people who haven't been asked to write it down.

That's it for this week. Talk soon.

— Kevin Kerner, CEO, Mighty & True

The Signal

New G2 data shows how B2B software discovery is changing (1 min read)

6sense analyzed 59,283 G2 listings across 218 software categories to show how company age and category maturity correlate with buyer discovery patterns, finding that traditional search-based discovery is giving way to other paths to vendor consideration. The research maps how long it takes software companies to get discovered and how that timeline has shifted.

Why it matters: If your demand gen plan still assumes buyers find you through search, this data shows you need a new map of how your category actually gets discovered.

What 63 million tool calls reveal about how AI agents actually behave (1 min read)

PostHog analyzed 63 million MCP tool calls to uncover patterns in how AI agents actually select and use tools in production. The research reveals common failure modes and usage patterns that differ from how agents are expected to behave.

Why it matters: If your team is building or deploying AI agents for marketing workflows, this data shows you where agent tool-use breaks down before you hit the same failures in production.

Tools & Tactics

How Anthropic's growth team decides what to build versus buy (1 min read)

Humans of Martech interviews Anthropic's growth engineering lead on their internal martech stack, covering Terraform-managed marketing configs, shadow tool sprawl, and a practical framework for build vs buy decisions. Includes the 'Kallax test' and the dynamics of vendor negotiation standoffs.

Why it matters: If your team is fighting the same build-vs-buy battles with your martech stack, this gives you a concrete framework and vocabulary straight from a company scaling fast.

Inside the paid-UGC engine that scaled a $2B AI startup (1 min read)

Lenny interviews Hector Serrano, the marketer behind Wispr Flow's paid-UGC growth engine, on hiring young talent over polished resumes and why top marketers resist over-relying on AI tools.

Why it matters: If your growth team is stuck in traditional paid playbooks, this breaks down a real paid-UGC system that scaled a $2B AI company and the hiring bets behind it.

Resources & Tools

How does your team compare?

The Mighty and True self-assessment compares your team's AI operating model, content velocity, and pipeline systems against mid-market B2B tech peers. Takes 10 minutes. You get a scorecard with concrete weak points you can attack this quarter.

▶ Take the Self-Assessment

Want a second set of eyes?

Running a marketing team right now feels like rebuilding the plane mid-flight. If you want a second set of eyes on what is actually working (and what you can safely cut), book a free, 45-minute Blueprint session with an expert advisor. We'll come back with the fix and a 90-day plan to deliver it. No pitch deck, just an honest read.

▶ Book your Blueprint