SMX Munich 2026 is one of those conferences where you walk in with a handful of questions and leave with a longer list of them. That’s not a complaint — it’s genuinely the point.

Two days, seven parallel tracks, 65+ sessions, 90+ speakers, and the kind of hallway conversations that often end up being just as useful as anything on the agenda.

Our own Fred Vallaeys, Founder & CEO, and Aaron Levy, Evangelist, were both at SMX this year. Fred ran the Advanced Google Ads Workshop on Day 0 and conducted a two-day workshop on vibe coding, while Aaron led a session diving into what’s actually working with AI, and what isn’t. So, suffice to say, we were paying close attention.

Here’s our honest read on the most important themes that came out of the conference.

The marketing bottleneck isn’t the tech, but the adoption

If there was one thread running through almost every session at Munich, it was this: the bottleneck in marketing right now isn’t the technology itself, but its adoption. More specifically, it’s the gap between what AI tools can do in a demo and what they actually do when you deploy them in a real workflow, with real stakes, and real people who need to trust the output before they act on it.

Boaz Ashkenazy’s opening keynote made this concrete. As the Senior Director of AI Infrastructure at Redapt and drawing over his years of experience with enterprise AI deployments across legal, medical, and finance, he argued that most AI tools fail not because they don’t work, but because they don’t earn trust. The formula he kept coming back to was simple: for a tool to actually get used, it has to provide relief from tedious work and increase the user’s sense of agency.

As he put it, “trust is not a feature you tack on at the end; it actually is the product.”

This trust problem shows up in PPC accounts just as much as it does in enterprise software deployments. Sam Tomlinson, Executive VP at Warschawski, made the career version of this argument in his session on evolving from campaign manager to, what he called, the “Growth Architect.”

The tactical skills that defined a strong PPC practitioner five years ago — bid management, keyword architecture, match type strategy — are now largely handled by the platform, and competing on those skills alone is a diminishing game. The work that remains is deciding what the machine optimizes for, what data it gets to see, how the creative ecosystem is built, and where the guardrails go. That kind of judgment is harder to learn from a certification and a lot harder to automate.

Matt Beswick, Co-Founder of aira, pushed this further in his session on future-proofing the PPC career. The traditional T-shaped marketer, with deep expertise in one channel and awareness of others, is genuinely at risk now because AI closes knowledge gaps faster than people can develop them.

What it still can’t do is connect marketing activity to real business outcomes, exercise judgment when the data is ambiguous, or show empathy to maintain client relationships. Those are the things worth doubling down on, even with automated systems in place.

Fred brought vibe coding to the main stage

Frederick Vallaey’s talk on “Vibe Coding: How Marketers Build Tools That Build Themselves” was one of the more practical sessions of the conference (if we can say so ourselves!).

The premise is that using natural-language prompts to generate working software, via tools like Lovable, has removed the engineering bottleneck in building custom marketing tools. What used to require a detailed spec, weeks of developer time, and a long debugging cycle can now happen in a single evening of prompting.

As Fred put it, “Software is no longer something you buy. It’s something you imagine, and it builds it just for what you need.”

He made this tangible by walking through things he’d actually built in the weeks before and during the conference.

• The first was a conference note-sharing app at notedtalk.com, built in roughly six hours after landing in Germany. Igiv mt lets attendees record audio notes, mark key moments, and automatically generate blog posts from session transcripts. [This blog was drafted with the info collected on notedtalk.com as well!]
• The second was an SEO audit tool he put together after catching a detailed AI-era SEO prompt on a speaker’s slide during a keynote. He photographed it, fed it to an AI, and had a working web app analyzing domains from an AI-search perspective by lunchtime.
• He also showed a PPC Advent Calendar from a recent holiday campaign, complete with daily reveals, sponsored prizes, and a PPC-themed Wordle variant, where the legal review for the prize giveaway actually took longer than building the site itself.
• Other was an internal social media ad generator that combines a product image, brand guidelines, and obscure daily holidays to produce on-brand posts automatically.
• And of course, our very own Optmyzr’s Sidekick, a chat-based AI assistant that lives inside your account, and can diagnose complex performance issues (like PMax cannibalization or budget constraints), identify hidden optimization opportunities (like isolated top-performing assets or high-spend waste), and instantly turn them into fully drafted, ready-to-run strategies and automation.

The principles behind the approach are worth reading in full in Fred’s LinkedIn post, but the short version is this: describe the outcome you want rather than the technical method, know where the risk line is, and connect vibe-coded tools to platforms with existing business logic rather than trying to recreate that logic from scratch.

The reason this matters beyond the novelty is the question Fred ended with: instead of asking how you complete a task, start asking whether you could spend thirty minutes building a tool that does it for you permanently.

Fred also hosted a two-part hands-on Deep Dive on Day 2, where pre-registered attendees went from prompt to working product during the session itself. A few people walked out with tools they were planning to actually use the following week.

PMax is still complicated, still infuriating, and still worth it

This ties directly to the discussion about setup and guardrails from earlier sessions. Mike Ryan, Head of Ecommerce Insights at Smarter Ecommerce, delivered a data-intensive presentation on Google’s Power Pack. He emphasized that the way you structure your campaigns is more important than many practitioners realize.

Mike pulled from an analysis of over 41 billion impressions across PMax, Demand Gen, and AI Max. As PMax revenue growth plateaued, Google introduced a multi-campaign approach to cover different funnel stages:

• PMax for direct performance,
• Demand Gen for mid-funnel across YouTube, Discover, and Gmail, and
• AI Max layering synthetic keyword targeting on top of standard search to catch the conversational queries that traditional keywords increasingly miss.

In theory, they complement each other. In practice, there’s significant overlap between PMax and Demand Gen on placements like Discover and Gmail, and Dynamic Search Ads-style technology now runs across all three simultaneously, which makes attribution genuinely difficult to untangle.

The self-competition issue in hybrid PMax and Standard Shopping setups came through clearly in his data. Running both in parallel tends to inflate CPCs, and Google’s official position that the campaigns don’t bid against each other doesn’t hold up under account-level analysis.

His recommendation was to use Demand Gen for top and mid-funnel work and keep it away from conversion-ready placements where it competes with PMax.

That lack of control is a real frustration, and the Global State of PPC 2026 data confirmed it isn’t just a vocal minority.

During their live session at SMX Munich, Fred and Wijnand Meijer, CEO of TrueClicks, unpacked the data from 1,306 practitioners, revealing that a staggering 53% of professionals now find managing Google Ads harder than it was two years ago.

This “complexity creep” is primarily driven by the same automated black boxes Mike Ryan analyzed, with 48% of practitioners citing a lack of granular control as their single biggest PMax frustration. To enforce the guardrails Mike discussed, the report found that 37% of advertisers are now running hybrid setups (Standard Shopping alongside PMax) to maintain manual oversight, while 17% have resorted to feed-only builds specifically to restrict where their ads appear.

Yet, in a bizarre twist, 65% of respondents still reported being satisfied with their results overall. It suggests that while the setup is infuriating and the attribution is a mess, the performance—when structured carefully—is still high enough to make the struggle worth it.

“Measurement theater” needs to end

Once you accept that the job is less about pulling levers and more about designing the system, the measurement question becomes unavoidable. If you’re not measuring the right things, you don’t actually know whether the system is working. Two sessions at SMX Munich 2026 made this case directly, and together they’re worth treating as a pair.

Jono Alderson, award-winning technical SEO consultant, made this case directly. Rankings, traffic, and platform-reported conversions have become what he called “measurement theater” — they measure how a campaign performs inside the platform, not whether marketing is generating anything real for the business. As AI intermediaries handle more query types and synthesize answers before users ever click, those metrics get weaker as proxies for actual business outcomes.

His proposed replacement was a six-layer measurement hierarchy, starting with technical and experience fundamentals and moving up through brand distinctiveness and reputation stability. His point was that optimizing isolated metrics is like polishing a single tile without ever stepping back to look at the mosaic.

Sam Tomlinson’s session reinforced this shift, advocating for a transition from average to incremental measurement. He argued that reliance on vanity metrics like traffic and rankings fails as AI intermediaries increasingly synthesize answers before users click, weakening traditional proxies for business value.

To counter this, he proposed a measurement hierarchy, beginning with technical fundamentals and ascending to brand distinctiveness and reputation stability. With accessible tools like geo-holdouts and open-source models such as Meta’s Robyn or Google’s LightweightMMM, the barrier to this rigor is no longer technical capability, but the practitioner’s willingness to exercise high-level judgment over automated systems.

This is exactly where the session by Inderpaul Rai, Director of Paid Media at WeDiscover, came in. If Alderson and Tomlinson were emphasizing about what to measure, Rai’s session stressed on what to feed the machine in the first place.

His concept of the Cobra Effect is worth understanding: it’s what happens when an algorithm is given a specific goal but achieves it in a way that quietly hurts the business. In marketplace businesses during supply shortages, for example, internal prices spike, leading to a handful of high-value conversions. Google reads these as wins and responds by aggressively hiking bids, creating a feedback loop in which you’re paying inflated CPCs for revenue that would have happened anyway.

Inderpaul shared a case study in which his team intentionally capped revenue outliers in conversion values to correct for algorithmic bias, a move that “tricked” the bidding system into being more conservative, resulting in a 9% CPC reduction and higher ROAS without actual revenue loss.

The principle behind it is that your job is increasingly about curating the signals the algorithm receives, not just running campaigns. What you choose not to tell Google can matter as much as what you do.

Automation is only as good as what you put into it

Aaron Levy, Brand Evangelist at Optmyzr, ran in the SMX Partner Track on Day 1 with a session called “New Best Practices: What’s Working (and not) in the Age of AI.” The pattern he kept coming back to was this: automation performs best when someone has made deliberate decisions about structure and signals before the campaign goes live—not handed over the keys, not trusted the default settings.

The accounts that struggle with Smart Bidding or Performance Max almost always have the same underlying problem: conversion data that isn’t clean, campaign structure that doesn’t reflect how the business actually makes money, or optimization targets that weren’t thought through carefully before being handed to an algorithm.

The automation isn’t the problem in most of these cases. The setup is. That’s a less comfortable answer than blaming the platform, but it’s the more accurate one.

Wil Reynolds, Founder and co-CEO of Seer Interactive, made the same argument in his closing keynote, just at the brand level. The race to produce AI-assisted content faster has created a massive volume of interchangeable, low-trust output, and his view was blunt about it.

The brands that win over the next decade won’t be the fastest at producing forgettable content; they’ll be the ones that built something genuinely hard to replicate. His challenge to agencies specifically was binary: invest in the creative talent required to do differentiated work, or accept being commoditized by the same tools you’re using to cut costs.

The honest takeaway

The message, across sessions after sessions, is this: the tools are genuinely powerful, the automation works when you set it up properly, and the practitioners who are struggling are mostly because they haven’t updated how they think about their role.

That’s the hard part. It’s not learning a new platform or a new feature. It’s recognizing that the job is less about doing things inside ad platforms and more about deciding what those platforms should do — and having the business understanding to make that call well.

For what it’s worth, that’s been Optmyzr’s position for a while now. Automation layering isn’t about giving up control. It’s about being deliberate about where human judgment adds value and where it doesn’t. SMX Munich 2026 was a conference built around exactly that question.