When Sam Isn't There to Check
Kraftisms #26 - In life - and in agentic advertising - everything we do and say is based on an assumption... that someone is there.
Quote(s) of the Week
“I want to hold your hand.” -John Lennon & Paul McCartney
“Under pressure, you don't rise to the occasion. You sink to the level of your training.” -U.S. Navy SEAL training aphorism
Kraftism of the Week: Curling Injury Alert
I tore my wrist’s all-important scapholunate ligament while curling.
Yes. Really. On ice.
For the record, it wasn’t a “double touch” or me touching the granite with an outstretched finger like we saw during the Olympics. It was a delivery. Full lunge, full extension, full release on sheet ice. The stone went exactly where I wanted it. My wrist did not. Surgery was Monday.
My dominant right hand has been in a splint all week. I’ve been dictating, hunt-and-pecking, wincing every time I try to text, and discovering something I genuinely didn’t expect: every system I thought I had was quietly running on infrastructure I never thought about. The fast reply sent mid-meeting. The contract redlined in minutes. The phone carried in my right hand while talking on it. None of those were skills I built. They were assumptions I inherited. The hand was always there, so I built everything around it.
You don’t have to tear a ligament to feel this.
Put your dominant hand behind your back right now. Try to type. Try to open a door. Try to do a pushup.
That last one is the point. I can’t do a one-handed pushup right now - and if I’m honest with myself, the surgery isn’t the real reason. The real reason is that I never trained for one. I built twenty years of upper body capacity on the assumption that both arms would always be available. When the hand went away, the pushup was no longer going to happen.
Applying the Kraftism: Skills Require Infrastructure
I’ve been thinking about that all week. Because there is a version of exactly this problem sitting at the center of the agentic advertising conversation right now, and I don’t think enough people are naming it directly.
Alan Chapell wrote about it in Marketecture this week in a piece he called “Agentic’s Big Gaping Hole.” His framing: agentic is being positioned as the thing that finally fixes programmatic’s endemic problems - the opacity, the misaligned incentives, the take rates that appear only when someone bothers to look. But agentic at machine speed, without compliance infrastructure underneath it, doesn’t fix those problems. It automates them. Alan identified the gap clearly. I want to build on what he started.
The Compliance Pushup
I keep coming back to the same thing: the vetting that kept programmatic from going fully sideways was never really in the system. It was in people. The trading desk lead who kept a mental list of SSPs with fill fraud issues. The media director who knew which data vendor had regulatory exposure. The programmatic manager who called someone at the DSP at 2am because a campaign looked wrong. None of that institutional knowledge lived in an allowlist or blocklist. It lived in the muscle memory of experienced practitioners who spent years being in the room. You know who I mean. Every company has at least one. Their name is probably Sam.
That is the dominant hand. Nobody had to build a backup for it, because it was always there.
Even when we talk about automation and agentic… the very words we use (compliance, risk mitigation, trust) are artifacts of a world where we couldn’t imagine a system that didn’t have the human in the chain. We may not even know the right questions to ask yet.
Chapell outlines five compliance questions any agentic transaction should be able to answer - privacy policy presentation, data flow integrity, consumer choice signal respect, targeting technique safety, and special category triggers like political or children’s content. All five are right. Here are the ones I’d add or expand.
What data is actually being used, and did a human vet it - or did the agent assemble an audience segment from a model that was never audited against a real population? “AI-derived” and “accurate” are not synonyms. I have sat in enough data systems to know the delta between those two things can be significant, and at machine speed that delta leads to enormous gaps.
Who is on the other end of this transaction, and can the agent actually verify that entity is who it says it is? Agent identity is still largely self-declared. The infrastructure for cryptographic verification is being built. It is not finished - and this applies equally to the seller registering inventory and the buyer activating against it.
Who is the named, reachable human who is accountable when (not if) something goes wrong: a brand safety violation, a misfire, an audience activation that shouldn’t have happened? I wrote about the principal-agent problem in Issue #25. This is the infrastructure version of that same question. Accountability without a face attached is an illusion, not a system.
Is the fee structure visible end to end? Programmatic has a long and well-documented history of fees that appear only under audit. Agents operating at machine speed without fee transparency baked into the protocol will inherit that history, not escape it. This is one of the key points that has to change if we want to avoid just automating the mistakes of the past.
Is the creative running in a context that a human would approve if they saw it before it ran? Context-aware placement is hard enough for people with dashboards. Agents need guardrails here, not inherited assumptions.
These aren’t theoretical concerns. They are the informal compliance checks that two decades of practitioners have been doing manually, often without realizing it. We never built the one-handed pushup version. Because we never had to.
Who’s Building the Infrastructure
This is not a slow-down argument. The one-handed pushup infrastructure for agentic advertising is being built right now, by people thinking seriously about exactly these problems.
Two efforts deserve your attention. They are being positioned as in competition with each other - by themselves and others. I’m still on the fence on whether they actually have to be… but I have a lot more perspectives to hear through my Listening Tour meetings to go before I can truly answer that question.
IAB Tech Lab’s AAMP - Agentic Advertising Management Protocols - extends the standards the industry already runs on: OpenRTB, TCF, etc. Its three-tier architecture covers Agentic Foundations, Protocols, and a Trust and Transparency layer that includes Agentic Guardrails - sandboxing, cryptographic verification, intent declarations - still in active development. Tech Lab’s Tony Katsur is clear that this is a marathon, not a sprint: “The promise of agentic AI is real and meaningful, yet its practical application will require years of market experimentation, standardization and alignment across platforms, agencies and publishers.”
AAO’s AdCP - the Ad Context Protocol, governed by the Agentic Advertising Organization - approaches the same problem from a different direction. Agent-native from the ground up, AdCP is designed as the allocation and reasoning layer across the full advertising market: a shared language for buyer and seller agents to communicate across inventory, audiences, and campaign objectives, built on open standards. AdCP co-founder Brian O’Kelley put it this way at launch: “It feels to me very similar to the early days of programmatic, when entrepreneurs were building foundational companies.”
They are building toward the same thing: agentic advertising that can execute at machine speed without leaving compliance running on a hand that might not always be available. That will not be quick… but I encourage everyone to get involved so you aren’t left behind.
To Wrap It All Up
Here’s the practical move. Take Alan’s five questions. Add the ones above. Run them against whatever agentic evaluation you’re currently doing - vendor, pilot, protocol, RFP. For each one, ask: if a human weren’t in the loop, would this answer still hold? If no, you’ve found your gap. If “I’m not sure,” that’s the same as no.
The training for the future looks like this: map your current compliance checks to the people who do them, not the systems that appear to do them. Find the ones where the honest answer is “we call Sam.” Sam is your dominant hand. Sam may not be in the loop when agents are transacting at scale… or you may find that Sam is irreplaceable.
AAMP and AdCP are both further along than most practitioners realize. The one-handed pushup infrastructure is being built. But compliance capability - like upper body strength - doesn’t appear the day you need it. It comes from work done before the injury.
Start training now. Before you find out which hand your whole operation was counting on… because it’s no longer there.
Andrew