Where should AI agents stop and human judgement begin? | Tech Radar

Overview

News, deals, reviews, guides and more on the newest computing gadgets

Start exploring exclusive deals, expert advice and more

Details

Unlock and manage exclusive Techradar member rewards.

Unlock instant access to exclusive member features.

Get full access to premium articles, exclusive features and a growing list of member rewards.

Where should AI agents stop and human judgement begin?

AI agents can optimize campaigns, but humans provide essential oversight

When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works.

AI tools in advertising have largely worked the same way for the past few years. They analyze campaign data, flag what is underperforming, and tell you what to do about it. The actual doing of logging into the platform, finding the campaign, making the change - still sits with the advertiser.

That is starting to change with a growing number of advanced platforms now letting AI agents execute campaign changes directly. Now, you can tell the agent to pause all push campaigns with a CTR below 0.3% and raise bids by 15% on the top three performers, and it will do it. No dashboard, no manual steps. The agent acts for you, which is useful in saving time on the manual work, so advertisers can focus on strategy, testing, and scaling instead.

Leading Ad Techs are upgrading their AI tools, with agents or ‘campaign co-pilots’ to support advertisers with campaign creation, editing, targeting, budgeting, scheduling, creative management, and reporting capabilities all through a single conversation, rather than working through a series of dashboard setup forms.

However, AI agents thrive when advertisers set them up for success with the best data, context and strategy to learn from, and act on. Without advertisers’ critical input and oversight, there are potential complications and AI use becomes counter-intuitive, opening up real implications for the industry.

So how can advertisers effectively work with AI? And where is the line between AI agents and human judgement?

From experimentation to execution: why AI in B2B marketing must now prove commercial value

Why AI coding agents keep stalling before production and the governance controls that fix it

The difference between AI that recommends, and AI that executes, is where the accountability sits.

For example, when a human reviews a recommendation and makes a change, the decision is theirs, and if it goes wrong, you can trace the reasoning. When an agent executes autonomously, that chain is less clear.

An automated bid increase applied at scale might look right based on the data, but the data does not know about the competitor announcement that went out that morning, or the internal brief that changed the campaign's priorities, or the brand issue being handled in the background. A human would have caught any of those. An agent running on last night's data would not.

This is not a reason to avoid execution-level AI. It is a reason to be specific about where it is and is not appropriate.

There is a clear category of campaign tasks that are well suited to autonomous execution such as pausing campaigns that hit their budget caps or generating performance summaries. These are operational tasks. The strategic decision has already been made, and the agent is now just carrying it out.

How AI agents will change how people work — and what they need from a PC

Protecting creative storytelling in an AI-first marketing world

AI agents aren’t the end of Saa S – they’re driving its next phase of growth

The data also suggests that these tools work significantly better when given proper context. Testing across agentic campaign setups has shown that advertisers who share detailed information about their goals, funnel structure and target CPA see substantially better outcomes than those who keep instructions minimal - in some cases the difference runs to over 100% in conversion performance.

The agent performs better the more it understands about what you are actually trying to achieve.

Alongside what agents should be allowed to do, there is the question of how they get access in the first place.

The industry is moving toward MCP-based integrations, a protocol that lets external AI agents connect directly to ad platform APIs. Rather than logging into a platform's own interface, the advertiser works inside whichever AI environment they already use, and the platform becomes something the agent calls when it needs to act.

Before that access is shared across a team or handed to a third party, the scope of the token needs to be thought through carefully.

The broader shift toward interoperability is probably the right direction. But there is a difference between an agent that can read your campaign data and one that can change it. Execution-level access needs to be more deliberate than the governance around reporting access, because the risk has now moved from what the agent can see to what it can do.

The honest answer to where the line should sit between autonomous execution and human oversight is that it depends on how clearly the strategy above the automation has been defined. Most campaign setups currently rely on a human being in the loop to fill the gaps.

That might mean spotting when a threshold no longer reflects the campaign's actual goals or when something happening outside the platform should change what is happening inside it. Remove the human from the loop, and those gaps turn into errors.

Before handing execution to an agent, it is worth taking the time to work through some basic questions. Which conversion events actually matter to the business, not just which ones are easiest to track? What does a routine adjustment look like versus a decision that needs a human sign-off? When should the agent flag something rather than act on it?

These are not complicated questions, but most campaign setups have never needed to answer them explicitly. There has always been a human available to exercise judgment in the moment. Agentic AI makes that implicit judgment into explicit rules, and getting those rules right before the automation is running is far easier than fixing it after.

We already know that AI agents can act. The harder question is whether advertisers, platforms and partners are ready to define what they should be allowed to act on. Execution-level AI could make campaign management faster and less manual, but it will only work if the rules around it are clear from the start.

That means setting limits, giving agents enough context, and keeping human judgment close to the decisions that still need it. The point is not to hand over control for the sake of it. It is to be much more deliberate about where automation genuinely helps, and where a person still needs to make the call.

This article was produced as part of Tech Radar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of Tech Radar Pro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

You must confirm your public display name before commenting

Surfshark gives its i Phone VPN app a cleaner look, and the browser extension gets one too

The Dreame X60 Pro Ultra Complete is a ridiculously clever and exceedingly competent premium hybrid robot vac for the well-heeled

I wasn’t a fan of Silent Hill f, but Silent Hill: Townfall has made me believe in the horror series again

Lanterns episode 6 introduces a character with major comic book ties to Hal Jordan — and it's someone we first met in Ryan Reynolds' disastrous 2011 Green Lantern movie

Quordle hints and answers for Monday, September 21 (game #1701)

Tech Radar is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site.

Key Takeaways

News, deals, reviews, guides and more on the newest computing gadgets
Start exploring exclusive deals, expert advice and more
Unlock and manage exclusive Techradar member rewards
Unlock instant access to exclusive member features
Get full access to premium articles, exclusive features and a growing list of member rewards