Kevin's Take
30 AI Agents in Production: What Actually Breaks
One crafty SaaS operator just published what might be one of the most useful AI article In his post, he shared what happened when you deploy 30 AI agents across their GTM stack.
I'm interested in this for our own business, but also sharing it because most of you are in the middle of this exact experiment right now... or about to be, and the gap between the demo and production is pretty big.
The Five Issues I'm Seeing
First, agent sprawl is a real thing. I'm kind of experiencing this now. In the article, this person started with three agents, then twelve, then thirty. Every team was then spinning up their own solution. Within six months, they had agents doing overlapping work, contradicting each other, and zero governance framework.
Second, the maintenance burden is being wildly underestimated. These agents need constant tuning. Prompts drift. Data sources and credentials change. What worked in February breaks in March. For example, one of their SDR agents started hallucinating product features after a website update. They caught it in QA. Barely.
Third, version control becomes a nightmare. When you update an agent's instructions, you're essentially changing its behavior. But unlike code, there's no clear rollback. No diff tool. Just vibes and testing. This can be managed with GitHub, but it needs careful setup.
Fourth, cost monitoring is broken. Most teams have no idea what their agents actually cost to run. Token usage varies wildly based on complexity and input length. Their content summary agent was spending 40 dollars a day because someone fed it unfiltered CRM notes. For Mighty & True, we have a daily Slack channel that monitors our token usage across keys in use.
Fifth, the human handoff is still terrible. Every agent eventually needs to escalate to a human. The article shows how messy that transition gets context loss, duplicate work, frustrated customers who feel like they're starting over.
What This Means for Your Q2 Planning
If you're budgeting AI initiatives right now, triple your estimates for ongoing maintenance and human oversight. The build is 30 percent of the work. The keeping-it-working is 70 percent. The good news is even with this added cost budgeting, the numbers are small. Automations are pennies where manual work is dollars (sometimes thousands of dollars).
Don't let teams spin up agents in isolation. You need a central registry, shared evaluation criteria, and someone who owns the strategy. Otherwise you're building technical debt that looks like innovation. We do this at Mighty & True by investing in AI and Automation Engineers that manage the build, maintenance and feature updates to all our AI code. It's a game changer.
And before you deploy anything customer-facing, build the failure mode first. What happens when the agent gets it wrong? How does a human take over? Can they see what the agent already said?
The Signal
Demand Gen Report's 2026 B2B Trends Research Report is Live — Demand Gen Report
Original research surveying 300-plus B2B marketers just dropped, and AI dominates the narrative. The report details how marketers are actually using AI, where they're seeing ROI, and where they're still struggling to move past pilot programs.
Why it matters:This is benchmarking data you can take to your board when they ask why you're not moving faster on AI or why you're moving at all. The gap between what marketers say they're doing with AI and what they're actually achieving is the story of 2026 so far. Use this to pressure-test your own strategy and see where you're ahead or behind the curve.
Clutch: 25% of Content Teams Now Target LLMs as Their Primary Audience
A new survey of 459 marketers from Clutch and Conductor found that nearly a quarter of teams now say LLMs, not humans, not traditional search, are their primary content audience. That number jumps to 32% at enterprises with 500+ employees. Meanwhile, 87% expect content budgets to increase in 2026, with reputation management as the top goal for both marketers and leadership.
Why it matters: The budget increase isn't surprising. Everyone's spending more on content. What's interesting is where they're pointing it. A quarter of teams have already reoriented their content strategy around machines, not people. If you're a CMO still thinking about content as "blog posts that rank on Google," you're optimizing for yesterday's discovery model. The teams pulling ahead are designing content systems that perform across humans, traditional search, and LLM-driven answers simultaneously.
On Our Radar
OpenAI's COO just said the quiet part out loud: AI has not yet really penetrated enterprise business processes. If you're feeling behind on AI, you're not. The gap between hype and production deployment is still enormous, even at the companies building the tools. Worth reading if you're managing stakeholder expectations around AI transformation timelines.
The Knot is one of the first brands running ads inside ChatGPT, and they're figuring out the rules as they go. No playbook yet. No benchmarks. Just experimentation. If you want a case study on what early adoption actually looks like in a new channel, this is it.
Apollo launched a native connector inside Claude via MCP. Sales teams can now search prospects, enrich contacts, and launch sequences without leaving the conversation. This is what "agentic GTM" looks like when it stops being a slide deck.
FROM THE TRENCHES
The Execution Gap Is Real. And Marketing Leaders Are Talking About It Out Loud.
We hosted 20 tech marketing leaders for a private dinner in Dallas last week as part of our Tech Marketing Rewired dinner series. No slides. No pitching. Just one long table and an honest conversation about what's actually happening inside their teams right now.
The pattern was impossible to miss.
Here's what we heard:
Everyone in the room agreed AI matters. That's not the problem anymore. The problem is that most teams are stuck somewhere between "we know we need to use AI" and "we're actually doing something with it." One guest shared a breakdown from their own org: about 10% of their marketers are truly leading with AI, 30% are experimenting, and 60% only touch it when someone tells them to. The trait that separated the top performers wasn't technical skill. It was impatience. They move fast, don't wait for permission, and teach by doing. Not by sending a Loom video.
The room also got into a real debate about brand. When AI can replicate your product in weeks, what actually differentiates you? The consensus: it's not features and it's not pricing. It's how your buyer feels about you before they ever talk to a salesperson. One guest put it simply. The number one reason deals are lost isn't price. It's confidence. That's an emotion. That's brand.
And then there was the conversation about GEO, Generative Engine Optimization. Someone brought up Resend, an email provider that has essentially structured their entire website for AI agents, not humans. And it's working. They're getting recommended by Claude, ChatGPT, and Cursor constantly. The stat that got the room's attention: roughly 85% of AI brand recommendations come from earned media. Not your website. Reddit threads, G2 reviews, press coverage. Your earned media strategy is your GEO strategy.
Why this matters for your team:
These dinners keep reinforcing something we see with our clients every week. The teams pulling ahead aren't the ones with the biggest budgets or the most tools. They're the ones who picked a real problem, got uncomfortable, and started building with an eye towards outcomes (not output). They stopped waiting for a perfect plan and started learning by doing.
That's exactly the transition we help teams navigate at M&T. They come to us knowing AI is critical but unsure what to build first, how to structure it, or how to get their team on board. We help them design growth systems that actually run in production. Not theoretical ones. Real ones their teams can operate and improve over time.
These dinners are becoming one of the most valuable things we do. Not because we show up with answers, but because we show up and listen. We're expanding the series to more cities this year, including Chicago, New York, and the Bay Area.
If you want to be on the list, reach out. Space is always limited. That's the point.