Angus Dowie, ADvendio's Global Head of Sales, looks at why an intelligent guardrail engine is essential as commerce enters its agentic era, and why the winners will be those who master governance at scale…

In 2026, AI’s capabilities in ad and media sales are no longer in question. Autonomous agents can negotiate rates, check inventory, optimise campaign yield, and build orders with minimal human input. 

What’s still in question, however, is whether the industry can handle that much delegated authority responsibly. 

Gartner, for one, predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, not because the models fail, but because of escalating costs, unclear business value, and inadequate risk controls. Meanwhile, Forrester’s June 2026 research reveals most enterprise leaders report adopting agentic AI, but only a small minority have it running safely in meaningful production, rather than what the report deems “agentish chatbots” dressed in autonomous systems’ clothing. 

Speed without boundaries is a business liability. When AI agents execute commercial decisions at scale across complex omnichannel stacks, an ungoverned agent has the potential to optimise toward a flaw with frightening speed, because it doesn’t carry the same assumptions about behavior and responsibility that a human advertising executive does, even when it’s rationalising in a way that looks similar on the surface. 

Without intelligent guardrails, agentic AI can make decisions that erode yield to maximise fill rate, or misapply pricing tiers. At scale, those decisions can drain advertising budgets or damage a company’s reputation. 

As adoption becomes more widespread, the technology itself stops being a differentiator. What separates winners from the rest is who masters governance at scale. 

A deterministic guardrail engine, one with access to real-time, comprehensive enterprise data, lets ad agencies and media companies capture AI’s speed without inheriting its liability. In practice, that means agents physically cannot execute a trade below a set floor price, apply a discount code outside a pre-approved range, nor commit inventory that hasn’t been verified against the live feed.

However, this is not a limitation on the technology, it’s what makes it usable at scale. Now that distinction is starting to separate the media companies that can trust their AI, with real budgets, from the ones still finding out the hard way what happens when they can’t. 

Moving beyond passive workflow automation 

That governance gap matters because of what’s actually being automated right now. The advertising and media industry is embedding AI agents deeper into campaign management, optimisation, and media sales workflows every year, and the results are, by all accounts, impressive. 

But most of what’s being sold as autonomous today is still, functionally, passive workflow automation. 

Technology vendors are promoting a future where autonomous systems can manage campaigns with minimal human intervention. They promise faster execution, improved efficiency, and better performance through "hands-free optimisation."; and adoption has surged on the strength of that pitch. 

What it hasn’t solved is the industry’s most persistent problem: AI systems operating without clearly defined governance, business rules, and operational safeguards. 

Automated decisions made with appropriate oversight can misjudge campaign pacing, misallocate budget, mismanage inventory, or miss contractual obligations, and those mistakes tend to show up as expensive makegoods, strained client relationships, and eroded trust in both the technology and the organisation deploying it. 

The real opportunity lies not in removing humans from the process but in establishing governance frameworks that enable AI to operate responsibly within well-defined boundaries. As the pace of automation accelerates, organisations must ensure that control mechanisms evolve alongside AI capabilities. 

Ultimately, handing agents the authority to make autonomous decisions at scale, without first addressing that governance layer, is how errors compound at machine speed instead of human speed. 

Three common pitfalls with poorly governed agentic commerce 

The pressing need to address the challenges associated with achieving true agentic commerce centers around three fundamental flaws with ungoverned automation. 

First, AI agents that rely on incomplete, outdated, or disconnected data from CRM platforms, ad servers, and other operational systems can make inaccurate assumptions about inventory availability. This results in planning errors and campaign delivery issues.

The second is misaligned objectives. Without clear business constraints, AI agents may prioritise metrics such as fill rate or campaign volume while overlooking critical commercial goals, including profit margins, pricing discipline, and adherence to established rate cards.

Finally, autonomous decision-making can create significant financial and regulatory risks when AI applies unauthorised discounts, generates billing inconsistencies, or fails to account for tax and compliance requirements across multiple business entities and jurisdictions.

As the velocity of AI increases, it exacerbates the risks associated with these flaws that can quickly spiral in what is becoming known as the "Unsupervised Velocity Trap”. Layering agentic AI on top of fragmented legacy stacks and the result is exponential risk, in the form of double-bookings, inventory hallucination, and margin erosion, to name a few. 

Getting past that trap requires the rethinking of what a guardrail actually is. 

The power of a true "Guardrail Engine"

Most companies already have some form of guardrail in place. The problem is that trusting an AI agent to work autonomously with real commercial activity takes a far more comprehensive form of governance than a simple spending cap or a loosely written brief. Rather, it takes deterministic rules, budget ceilings, real-time margin protection, and human-in-the-loop (HTL) checkpoints, working together rather than isolated safety nets. 

Agents need to operate inside hard, unbreachable boundaries set by human revenue leaders. That’s the difference between a deterministic rule and a probabilistic guess. 

Profit and margin protection has to go further than a rigid price cap. Agents need real-time guidance that responds to internal and external pricing signals, built around logic that enforces floor prices even when negotiating with external buyer agents. 

HTL controls let AI carry the administrative load, drafting proposals, checking inventory, without going rogue. Human teams keep approval authority over high-stakes strategy. 

And one practical way to tighten all of this is to deploy a fleet of specialised agents rather than one general purpose assistant that’s hard to define or audit. Agents that specialise in sales, inventory, or finance are easier to configure precisely. 

This is the underlying logic of ADvendio's agentic infrastructure: deploying specialised agents across omnichannel workflows, governed by a unified Guardrail Engine that enforces business rules before a transaction ever hits the ledger.

The enterprise infrastructure requirement

None of this works without the right infrastructure underneath it. Agents need a unified source of truth and real-time data to make informed decisions, whether that’s about sellers, finance or inventory. 

But feeding sensitive commercial data into a public or general purpose model isn’t a reasonable trade for that visibility. 

Prioritising data sovereignty in enterprise infrastructure decisions solves that problem because agents can access performance metrics, client contracts, and yield rules without any of that data leaving a secure internal environment. 

ADvendio’s Revenue OS, for instance, was built natively on Salesforce, and unifies omnichannel inventory, sales, and accounting into one synchronised stream, giving agents a clean dataset to work from without putting security or data sovereignty at risk. 

Seen this way, investing in enterprise-grade software isn’t a nice to have alongside an AI strategy. It’s what makes the ROI of that strategy tangible, and what keeps compliance intact without exposing sensitive commercial data in the process. 

Scaling agentic AI without the velocity paradox 

True agentic commerce in advertising cannot rely on "set-it-and-forget-it" automation. Autonomy is a force multiplier, and that cuts both ways. Without guardrails, it multiplies undesirable outcomes just as efficiently as good ones, with real financial and reputational consequences attached. 

That’s what undermines autonomous commerce at scale. Not the technology itself. 

As pressure mounts across the industry to plough forward with AI initiatives to capture speed and efficiency gains, governance becomes a key differentiator. Now is the time to audit their AI readiness and governance structure.

To discover how ADvendio's Agentic Revenue OS empowers publishers and media companies to scale revenue safely with enterprise-grade guardrails, request a demo with the team here.