Best of LinkedIn: MarTech Insights CW 32/ 33

00:00:00: provided by Thomas Allgaier and Frennus based on the most relevant LinkedIn posts about MarTech in calendar weeks, thirty-two and thirty three.

00:00:08: Frenness is a B to be market research company that supports enterprise marketing teams in unlocking the full potential of their customer data with the help of AI.

00:00:17: you can find more info in the description.

00:00:18: yeah so imagine an agent right?

00:00:20: it's just quietly scanning your CRM in the background It notices major prospect is slipping away.

00:00:27: okay It analyzes their past behavior, writes a highly targeted email and then just fires off an outbound campaign.

00:00:35: Wow!

00:00:36: All without a single human ever logging into a software dashboard.

00:00:40: like we are officially passed the era of those cute highly governed AI pilots.

00:00:45: Yeah The training wheels were off.

00:00:46: Exactly they're totally us.

00:00:47: We're looking at actual autonomous AI execution happening in live production environments

00:00:52: right now.

00:00:52: And that is exactly the mission for this deep dive.

00:00:54: today, we are unpacking this massive structural shift That's happening in cross-the entire Martek landscape.

00:01:00: We're going to look at how agentic AI has finally you know getting his hands on the keyboard and How our underlying data infrastructure?

00:01:07: Is just completely breaking it having to re architect itself to support that.

00:01:11: Right because I have to.

00:01:12: Exactly and fundamentally we want to look At how the day-to-day job of a BDB marketer is changing Because of all this.

00:01:20: Yeah,

00:01:20: it's a huge shift and I think the biggest breakthrough that's unlocking All This.

00:01:24: Autonomous Execution is A technology That Is just Dominating The Conversation Right Now.

00:01:29: It'S Called MCP

00:01:30: Okay MCV

00:01:31: Model Context Protocol.

00:01:33: So If You Aren't Familiar With It Think Of MCP As Like A Universal USB-C Cable but for artificial intelligence.

00:01:40: I

00:01:40: like that analogy,

00:01:41: right?

00:01:41: It's an open standard.

00:01:43: it allows an AI agent to plug directly into your company's proprietary data and your enterprise tools And Martin Keane highlighted a massive milestone on this front recently.

00:01:52: Oh

00:01:52: what is that?

00:01:52: he noted that Salesforce has officially moved its MCP servers to general availability.

00:01:57: Okay So it's out in the wild

00:01:59: exactly.

00:02:00: and What this means in practice Is that AI agents can now access data completely headless.

00:02:04: Headless meaning no UI.

00:02:06: Right, No user interface.

00:02:08: the agent can just query your Salesforce database pull out insights and take direct action without needing a human to click around a screen.

00:02:16: The UI essentially becomes optional.

00:02:18: at this point

00:02:19: that is wild.

00:02:20: I mean i love the idea of an AI diagnosing a problem instantly but let's be real for second.

00:02:25: yeah We saw a really great breakdown from Varun Parmar and Akhande Davis.

00:02:29: And they were discussing how Adobe is rolling out their own MCP servers for Gen Studio, and Marketo Engage.

00:02:36: Right!

00:02:36: I saw that.

00:02:37: Yeah...and

00:02:38: they highlighted this specific use case.

00:02:40: A marketer can open up say Claude or ChatGPT.

00:02:44: Just ask why's my ad campaign tanking?

00:02:46: It just tells you

00:02:47: Exactly.

00:02:48: The agent reaches into Marketo pulls the customer records and hands you an answer in chat.

00:02:53: But I have to push back a little here.

00:02:55: We'd go for it half the time.

00:02:56: these large language models hallucinate.

00:02:58: basic math

00:02:59: Yeah, they really do so.

00:03:00: if an AI hallucinates The reason my ad is underperforming and i just trust it blindly?

00:03:07: I might accidentally pull the plug on A you know million dollar campaign.

00:03:10: that was actually working perfectly fine.

00:03:12: Oh

00:03:12: absolutely.

00:03:13: So did we really trust these systems to act directly on our data?

00:03:17: Well, that is the exact tension industry is wrestling with right now.

00:03:21: But the crazy thing is... The adoption is pushing forward anyway and it's extending way beyond just those massive enterprise platforms like Salesforce & Adobe.

00:03:31: Where else have we seen it?

00:03:33: Well, Majaji pointed out that clay and you know Clay is a platform A ton of B to be teams use for data enrichment.

00:03:39: right

00:03:39: love clay.

00:03:40: Yeah They have now added an MCP interface along with the command line Interface.

00:03:44: so clay is no longer just a place where you build fancy spreadsheet tables.

00:03:49: You can use plain English To instruct an agent to execute a really complex Outbound campaign

00:03:54: Just by typing it.

00:03:55: Yeah,

00:03:55: the agent builds a workflow.

00:03:57: It scores The inbound leads and it runs the entire sequence automatically.

00:04:01: Okay let's just pause on that.

00:04:03: Giving an AI agent direct read-and-write access to our databases via MCP And letting it run live campaigns That's a terrifying level of action.

00:04:13: It

00:04:13: is a little scary yeah?

00:04:14: Its less like hiring an eager intern and more Like giving an algorithm the master override To like your city's traffic light system.

00:04:22: That's

00:04:22: a good way to put it, right?

00:04:23: Because if the AI optimizes for traffic flow at one specific intersection It might accidentally cause a ten mile gridlock everywhere else simply because it doesn't understand The broader context of the entire City

00:04:34: exactly.

00:04:35: so how do organizations ensure these agents don't just completely Crash their workflows and operations.

00:04:41: Well,

00:04:41: the answer is by fundamentally changing The approval mechanism.

00:04:45: you don't take the algorithm offline right?

00:04:47: You just stop it from turning the light screen without your permission.

00:04:50: Okay.

00:04:50: So putting a human back in the loop yes

00:04:53: Michael Baye shared A perfect example of how this Is actually being deployed safely In the real world.

00:04:57: he looked at How aws worked with formula one to handle Their massive data operation.

00:05:02: oh f-one has so much data

00:05:04: Right, and F-One had this agonizing eighteen month backlog.

00:05:08: They were just trying to onboard twelve new fan data sources into their MarTech platform.

00:05:13: Wait!

00:05:13: Twelve sources was an eighteen month back log?

00:05:15: Yeah the projection was eighteen months because of all the manual mapping and compliance work.

00:05:20: but using agentic AI they cleared that entire backlog in a matter of weeks.

00:05:25: Weeks?!

00:05:25: That's insane

00:05:26: It is...but here's the kicker.

00:05:28: The secret wasn't full autonomy it was really strict.

00:05:32: AI proposes humans review framework.

00:05:35: Okay, so the AI drafts the blueprint but it isn't allowed to like actually pour the concrete?

00:05:40: Exactly!

00:05:41: The AI does like ninety-five percent of their really tedious manual work.

00:05:45: It maps out data schemas, writes ingestion pipelines and configures GDPR compliance policies.

00:05:51: All

00:05:51: heavy lifting.

00:05:53: But then...it stops and it submits a pull request.

00:05:56: Right, the developer term?

00:05:57: Yeah which is just software engineering terms for saying hey I made these changes please review them before we merge them into the live system.

00:06:05: So a human still has to final say

00:06:06: A Human Engineer has to officially approve that Request Before Anything Actually Goes Live.

00:06:12: And We Are Starting To See This Exact Philosophy Get Baked Directly Into Marketing Platforms Now

00:06:17: Like natively.

00:06:18: Yeah Larry Huck Noted That Adobe Workfront AI Collaborators Just Reach General Availability And that provides a shared system of record.

00:06:26: Oh, I see where this is going

00:06:27: right.

00:06:28: so whether you build your agent on Claude or co-pilot Or writer the agent routes its completed work into Workfront and The human review happens directly in that workflow before any campaign actually ships to the public.

00:06:41: okay?

00:06:41: That transition from Humans You know doing the work two humans reviewing the work Is really profound.

00:06:48: it

00:06:48: changes everything

00:06:49: It does but also creates a new compounding problem.

00:06:53: If AI can build workflows and write code, and spin up custom apps at this kind of blistering speed our tech stacks are going to become completely unmanageable.

00:07:03: Oh

00:07:03: absolutely!

00:07:04: We aren't just talking about a crowded MarTech stack anymore.

00:07:06: we're staring down the barrel of massive system governance crisis.

00:07:10: The sheer economics software creation have entirely inverted.

00:07:15: Seven Goats shared really stark warning that he introduced concept called

00:07:20: the hyper tail.

00:07:21: Okay, what is that?

00:07:22: So for the last decade we've thought of the Martek landscape as having you know maybe fifteen thousand sauce tools globally

00:07:29: which already feels like too many.

00:07:31: Right but today We have something called vibe coding

00:07:35: Vibe coding.

00:07:36: Yeah, it's where non-technical marketers just type what they want in plain English and the AI writes The underlying code to build a fully functioning app.

00:07:44: Wow because of that A small team can build a niche single purpose application In a matter Of hours.

00:07:50: so guts warns That the landscape is splitting.

00:07:52: We're looking at a near future with potentially one million AI-generated micro tools living inside enterprise environments.

00:08:00: A million microtools?

00:08:01: Yeah!

00:08:01: And let me guess, a marketer will use an AI agent to build a custom tool for like one specific webinar campaign.

00:08:09: they'll use it three weeks and then completely forget that even exists.

00:08:13: It just quietly lives on as shadow IT in the background forever.

00:08:16: That is exactly what happens.

00:08:18: And this perfectly illustrates the concept that Bill Habib calls, The Platform Tax.

00:08:22: There's a platform tax!

00:08:23: Yeah because everyone focuses on software licensing fees right?

00:08:28: But the real cost of MarTech today Is the organizational burnout.

00:08:32: It's the human toll Of trying to absorb integrate and train your teams On new technology way faster than as humanly possible.

00:08:41: Real quick.

00:08:42: If you are a marketing leader trying to make sense of this chaos and want to stay ahead of these shifts, Make sure hit subscribe on the deep dive so that we don't miss our future additions.

00:08:51: We track these underlying architectural changes constantly… so you won't have to!

00:08:55: Now looking at shadow IT risk….

00:08:57: This feels like we're completely ignoring recent history.

00:08:59: How so?

00:09:01: Well Gabe Larson put out massive red flag about it.

00:09:03: He pointed us in danger repeating the exact same sauce sprawl mistakes on steroids.

00:09:11: Oh, yeah!

00:09:12: Back then we bought a different point solution for every tiny problem until the stack became this giant disconnected monster.

00:09:19: now We're about to do it with AI agents which

00:09:22: is worse.

00:09:23: right you could easily end up With a company stacking thirty different single-purpose AI agents.

00:09:29: You know an outbound SDR agent?

00:09:31: A customer support agent or recruiting agent and They all have thirty separate memories, thirty disconnected dashboards and they absolutely do not talk to each other.

00:09:41: And when things don't talk to one another you just lose efficiency.

00:09:44: You create severe catastrophic security vulnerabilities.

00:09:48: Oh

00:09:49: I hadn't even thought about the security side.

00:09:50: It's bad.

00:09:51: Clark Barron broke down a recent highly publicized breach at Clue which is competitive intelligence vendor.

00:09:58: Okay...and it perfectly illustrates danger of this ungoverned sprawl we're talking.

00:10:03: The intruder didn't hack a firewall, they didn't ingest the password.

00:10:06: And how did it get in?

00:10:07: They walked right through his side door using a long-dormant API credential from an abandoned integration prototype.

00:10:13: A developer simply forgot to turn it off.

00:10:15: Wait... An integration that never even went live?

00:10:17: Exactly!

00:10:17: Just a test that got forgotten.

00:10:19: If you're a RevOps manager listening this now That should terrify you.

00:10:23: You probably have three or four of these dormant API connections sitting in your stack today just gathering dust.

00:10:29: At

00:10:29: least, three-or-four.

00:10:31: and the damage it caused at Clue was staggering.

00:10:34: The intruder used that DEG credential to harvest OOTH tokens And you can think of an oath token as a digital skeleton key.

00:10:43: It allows one application to access another without needing the user's password every single time.

00:10:48: Right, it just keeps the door open.

00:10:49: Yeah.

00:10:50: So with those skeleton keys The intruder gained deep access into Salesforce HubSpot and Slack environments across at least a dozen connected companies.

00:10:59: That is

00:11:00: brutal!

00:11:00: It gets worse.

00:11:01: Here is truly chilling detail from Barron's breakdown.

00:11:05: The affected companies couldn't even tell what data had been stolen.

00:11:08: Why

00:11:08: not?

00:11:08: Don't they have logs?

00:11:09: Well, Enterprise platforms will show you which external apps have been granted access.

00:11:14: But there is no visitor log recording what those API grants actually do once they are inside the system.

00:11:19: Oh wow Yeah The permissions were standing silent and entirely unmonitored.

00:11:24: but wait if that clue breach happened because of a forgotten side door And we're about to unleash a million microagents That are all trying to connect our systems.

00:11:34: How do we possibly secure that?

00:11:36: It's a

00:11:36: huge challenge.

00:11:37: We can't build a firewall around a million tiny tools!

00:11:41: The only logical solution is to stop moving the data round in first place, right?

00:11:46: You're hitting

00:11:46: the nail on your head... Like

00:11:48: if the data stays at one central location….

00:11:50: …the agents have come into the data.

00:11:52: rather than us piping the data out to all the agents.

00:11:55: That is the exact architectural pivot that industry's making right now.

00:11:59: Ed Pop shared an analogy from Subu Desirajou, who visualizes this beautifully.

00:12:04: He compares the entire customer-data ecosystem with a city water supply.

00:12:11: Think about the flow.

00:12:12: You start with a reservoir which represents your source systems like CRM or website.

00:12:17: Then the water moves into treatment plant.

00:12:19: That's where you do all your data transformation, de-dplication and cleaning.

00:12:22: Okay making it drinkable

00:12:24: Exactly.

00:12:25: Next are the pipes Representing API handoffs between different platforms.

00:12:29: And finally You have the tap The Tap.

00:12:31: is your reporting dashboard Or in todays world Your AI prompt?

00:12:35: And the problem is, most marketers only ever look at the tap.

00:12:38: They just turn it on and assume that water's clean.

00:12:40: Precisely!

00:12:41: But if data is dirty in the reservoir or treatment plant fails to deduplicate records those errors compound every single step down the pipes.

00:12:51: By the time you reach a tap You end up feeding poisoned water To your AI models.

00:12:56: An AI model hallucinating bad data at scale Is a complete disaster.

00:13:00: It will confidently send wrong pricing Your biggest client.

00:13:04: So to fix plumbing Basically, the market is aggressively reorganizing where The treatment plan actually lives.

00:13:10: Yes real long.

00:13:11: acre pointed out a massive architectural shift recently when Databricks introduced customer Lake.

00:13:16: that was a big deal.

00:13:17: Yeah They specifically positioned it as an agentic CDP or customer data platform.

00:13:22: And why does that matter?

00:13:23: because its signals?

00:13:23: That traditional cdp functionality Is being completely devoured by the core data warehouse layer.

00:13:29: AI requires Massive raw computing power to run its predictive models.

00:13:35: It doesn't want a filtered, rigid view of the data sitting in a separate marketing tool.

00:13:44: That

00:13:45: was three sixty.

00:13:54: Yeah,

00:13:55: they are completing a transition from offering a standalone marketing product into providing pure underlying infrastructure.

00:14:03: The data doesn't even move in to Salesforce anymore.

00:14:05: Let's pause on that because that is a huge shift.

00:14:08: How does sales force access the data if it doesn't actually move?

00:14:12: through something called a zero-copy architecture.

00:14:14: Zero copy?

00:14:15: Yeah, instead of physically copying a million customer records from your Snowflaper Databricks warehouse over into Salesforce which you know takes time creates latency and costs a tonne money right salesforce just points to the data where it already lives.

00:14:27: think of it like giving someone the call number for a library book Instead of photocopying all four hundred pages For them.

00:14:32: Oh that makes so much sense.

00:14:34: The data stays resident in the warehouse And the applications Just read It In Place.

00:14:39: Ravi Saragi from Unifor echoed this exact sentiment.

00:14:43: He argued that the era of traditional rules-based CDP is officially over, done!

00:14:49: Instead teams are deploying custom small language models and these run predictive simulations directly on top of unified zero copy data before a campaign ever even launches.

00:15:01: But even with this incredible architecture, we are still running into fundamental flaws With how the software industry actually operates.

00:15:08: Yeah The business side gets in a

00:15:09: way totally near obvious highlighted A fascinating kind of counterintuitive problem regarding How we store?

00:15:16: This unified data.

00:15:17: what did he say?

00:15:18: well Let's see you work In a sector within incredibly long consideration cycle like booking luxury travel or buying a new car sure the highest value.

00:15:26: Customers and those industries often take years to finally convert.

00:15:30: A family might browse your automotive site anonymously for say, eighteen months.

00:15:35: They're configuring different car models looking at colors before they finally walk into a dealership.

00:15:40: But

00:15:41: because legacy marketing platforms charge companies based on the total number of profiles they store teams are forced to expire and delete anonymous profiles after ninety days just To save money on licensing fees.

00:15:53: Oh wow!

00:15:54: It's structural paradox.

00:15:57: The billing model of the software is fundamentally breaking the AI's capability.

00:16:01: Exactly!

00:16:02: The licensing model is forcing organizations to artificially wipe the memory from their own AI

00:16:08: That it so counterproductive

00:16:10: It IS.

00:16:11: So when that family finally comes back on the website in month ten, actually buy a car... ...the AI treats them like a total stranger.

00:16:18: It starts offering them entry-level sedans instead of the luxury SUV they've been researching for nearly a year.

00:16:24: Your technology stack is literally punishing your most profitable customer behavior because you didn't want to pay a few extra cents to store their profile,

00:16:31: which brings us to critical realization.

00:16:34: if data's being centralized in these massive zero copy data clouds and AI agents are taking over actual execution campaigns through MCPs what does human marketer actually supposed

00:16:46: do?

00:16:47: If

00:16:47: the machine writes a copy, builds it's flow and pulls its list.

00:16:51: What is job description?

00:16:52: The job titles themselves are actively changing right now to reflect this new reality.

00:16:57: Mike Rizzo and Ashley Langford observe that marketing ops leaders increasingly abandoning the Ops title entirely.

00:17:04: They're being rebranded as GTM engineers or marketing engineers.

00:17:08: GTM

00:17:08: engineers, I like that!

00:17:09: Yeah they're no longer just administrators who reset passwords for an email tool.

00:17:14: They are taking architectural ownership of the entire go-to market strategy.

00:17:17: That's a massive elevation of the

00:17:19: role It really is.

00:17:20: And we also saw interesting insight from Fabio Canalesi.

00:17:24: He noted enterprise companies are drastically shifting how they hire for CRM roles.

00:17:29: Who so?

00:17:30: Instead building out these heavily layered teams junior, mid-level and senior staff to manage operations.

00:17:37: They are hiring a single highly capable senior individual contributor just one person because the AI is handling the heavy lifting of writing the code and building the integrations.

00:17:48: One strategic expert can effectively do The work of a four-person team.

00:17:52: but that strategic human expertise Is more critical than it has ever been.

00:17:56: for reasons?

00:17:57: That technology alone simply cannot solve.

00:17:59: explain that.

00:18:00: so Matthew McDonough diagnosed the root cause of this brilliantly.

00:18:05: He talked about the ontological layer of business data.

00:18:09: Ontological, like the philosophy term?

00:18:11: Yes

00:18:11: exactly!

00:18:12: Ontology is just a philosophical study.

00:18:14: how we categorize reality.

00:18:16: every single application in your tech stack models reality differently based on specific job it was built to do

00:18:22: Right.

00:18:22: so if you ask a marketing platform what a customer is It defines them by their email address and audience engagement.

00:18:29: But If You Ask A CRM It cares about the company account and opportunity stage.

00:18:34: And if you ask The Billing System, a customer is just an active subscription ID in an invoice.

00:18:40: Exactly!

00:18:40: Each of those views are entirely legitimate for its specific function but none represents the complete truth.

00:18:53: dump all those conflicting tables into a data warehouse, ask an AI to write a sequel.

00:18:58: join and expect it to magically make sense of your business.

00:19:02: It does know what you mean?

00:19:03: A

00:19:03: technical joint between two databases is ultimately an expression of an organizational disagreement.

00:19:09: before the technology can ever work human beings have to sit in her room navigate the internal company politics and strategically agree on a unified definition of reality.

00:19:19: What officially defines a churn risk?

00:19:22: When does an MQL actually become a

00:19:24: sequel?

00:19:25: Yeah.

00:19:25: The AI doesn't know your company politics, it cannot make those philosophical decisions for you

00:19:30: And human logic is the only thing that prevents these high-speed machines from doing very stupid things...very quickly

00:19:36: Absolutely.

00:19:37: Christopher Marriott shared perfect real world example of this.

00:19:40: He received harsh payment decline notice From a progressive insurance company.

00:19:44: Oh I hate them.

00:19:45: Right But the catch was The notice were sent three days after he had already logged into portal and corrected his payment information.

00:19:52: The system worked perfectly from a purely technical standpoint.

00:19:56: A rule triggered a message and the agent delivered it, but it failed entirely.

00:20:00: form customer experience standpoint because the system lacked fresh context.

00:20:04: Just didn't know he fixed that?

00:20:05: Exactly!

00:20:07: Message speed means absolutely nothing if it lacks human logic in real-time awareness.

00:20:12: Sending wrong messages at this speed of light just damages your brand faster

00:20:16: Which brings us to ultimate realization for marketing leaders.

00:20:20: today John Miller shared a quote from Kyle Lacy that perfectly encapsulates where this entire industry is heading.

00:20:26: Let's hear it.

00:20:27: He said, fast forward to year or two every single marketing team in the world including your competitors Is going to be running the exact same AI stack?

00:20:36: That's true.

00:20:36: They will all have the same agentic workflows The same foundational language models and the same capability To generate a thousand personalized emails a minute.

00:20:45: When everyone has the exact same tools, The tools completely stop being a competitive differentiator.

00:20:51: So if the tech stack is just to commodity what's left?

00:20:54: What does it actual mode

00:20:55: human taste Human judgment yeah knowing which ideas are actually worth making in first place and know what kind of messaging is gonna resonate emotionally with another human being.

00:21:07: When AI levels the playing field on execution, your taste becomes

00:21:18: around AI taking marketing jobs.

00:21:20: It

00:21:20: really does!

00:21:21: It isn't replacing the marketer, it's just stripping away robotic tasks and elevating truly creative strategic parts of work that humans are actually good at.

00:21:31: But before we wrap up there is one final provocative angle to all of this that every B-to-B professional needs to consider.

00:21:38: Okay, lay on me!

00:21:39: We've spent this entire deep dive talking about AI agents acting on behalf of the company right?

00:21:44: Marketing agents sales agent support agents yeah but what happens when the buyer gets their own agents?

00:21:50: oh wow we haven't even talked about the customer side

00:21:52: Right?

00:21:53: think about your perfectly architected MarTech stack.

00:21:57: What happens to it when you're potential customer uses there own AI shopping agent To go out Analyze thirty different vendors, read all your technical documentation, scrape your pricing and disqualify you before a human being ever even visits your website.

00:22:13: That is terrifying!

00:22:14: Your

00:22:14: analytics dashboard won't see a bounce...your CRM won't record lost opportunity.

00:22:20: Marketing might not just lose the customer in this new era.

00:22:22: If you aren't architecting your content for AI to AI communication, You might lose the very first point of contact entirely.

00:22:29: Wow It's an existential shift that every GTM engineer needs to be thinking about right now.

00:22:34: if you enjoyed this episode New episodes drop every two weeks.

00:22:37: Also check out our other editions on field marketing channel marketing and partner ecosystem ai and btb go-to market abm And social selling.

00:22:45: Thanks so much for joining us on this deep dive into the changing landscape.

00:22:49: It's a wild

00:22:54: time!