🌟 Editor's Note
Welcome to the bi-weekly newsletter which focuses on the AI topics that leaders need to know about. In this AI age, it’s not the knowledge of AI tools that sets you apart, but how well they can be integrated in the context of your business.
This requires a focus on your people and helping them through the change above any AI product you can buy.

The Biggest AI Risk Isn’t Hallucinations. It’s Human Complacency

In this newsletter, we've spent a lot of time focusing on how people in our organisations are adopting AI. This article focuses on what happens after they do - and it is a less comfortable story.

A joint study by researchers at Auckland University of Technology and the University of Massachusetts Boston has identified a phenomenon called “AI complacency”: not a failure to use AI, but a failure to check it. The researchers set out to understand why this complacency occurs: is it our cognitive biases for speed and ease or something else? What the authors of the paper found was that employees who feel low accountability for verifying AI output tend to simply accept it, including when it is wrong.

What makes this interesting is what drives complacency. It is not laziness. It is not a lack of AI skills. It is not even experience. What drives it is structural: when people do not feel personally responsible for checking the output, they stop checking it. And when the task is complex, the workload is high, or the work happens in a team setting, it gets worse.

Takeaways for Leaders

When employees know they will need to justify their use of AI output, errors drop significantly. The question to ask your team: does anyone feel personally responsible for checking what the AI produces before it reaches a customer, a client, or a decision?

Being busy makes it worse. The research also showed that high task load combined with complex tasks is where complacency spikes most sharply - exactly the conditions where people are most likely to reach for AI in the first place. If your team is stretched, the risk of unchecked AI output goes up, not down. Building AI into busy workflows without changing accountability structures is a risk that compounds with pressure.

In collaborative settings, people assume someone else is checking. When AI is embedded in those workflows, that assumption becomes dangerous. Clear ownership of AI output matters significantly more when the work is interdependent.

Deploying AI is the easy part. The harder and more important work is building the human infrastructure around it. AI does not make organisations smarter on its own.
The organisations that will get the most from it are the ones that invest as much in how humans work alongside it as they do in the technology itself, looking at workflows, team structures and performance expectations around it.

Click here for the link to the paper

The Ad Industry's Dirty Secret: AI Is Already Winning

A large-scale study of 4,633 real advertising campaigns on the advertising platform Taboola has produced one of the clearest pictures yet of how AI-generated ads actually perform in the wild. Across more than 369 million impressions and 2.5 million clicks, researchers found that AI-generated images match human-made ads on click-through rates.

The more interesting finding was the subtext: when consumers could not tell an ad was AI-generated, those ads outperformed human-made ones. When the AI origin was apparent, performance dropped significantly.

The irony is that consumers are not always right about which is which. Nearly 59% of AI-generated ads were rated as probably or definitely human-made by study participants. AI naturally produces sharper, clearer images with more prominent faces, which are qualities people associate with human-made ads rather than machine output.

^{Takeaways For Leaders}

The cost case for AI ad creation is already strong. The fact that performance parity exists, and can be exceeded, strengthens it further. There is practical guidance in here for what people respond to. For instance, prominent faces and sharp and clear images work really well. Heavy saturation of colours or work that was too polished had a negative impact.

Many advertisers have been experimenting with AI ads. A quasi-experimental study such as this gives robust insight as it is using data derived from regression analysis across thousands of real campaigns.

There are limitations: this study was skewed towards performance-based advertisers and looks at clicks rather than sales. But the direction it gives in how to produce effective ads using AI is clear.
And it should be noted that since this study was executed, (much of the data capture was in 2023) there have been significant advancements in the image generation models. And so it is likely that not only will the principles of the findings hold, but AI ads that follow the success guidelines are likely to perform even better.

Click here for the link to the paper.

We'll Work Alongside The Chatbots

The debate of whether AI will destroy jobs or create new ones is not yet settled.
A look at the historical impact of digital technology may give us some insight as to how this plays out.
Microsoft released Excel 5.0 in 1993, and its software that got rid of the need for whole areas of number crunching that were previously done manually. And yet, rather than reducing employment, if we look at the number of jobs in the accounting and bookkeeping fields, this rose by more than 70% over the following decades.

In the age of AI, we've seen several false starts as companies have tried to replace teams with machines and failed. Customer service has been under the spotlight especially where chatbots have been introduced to deal with customer queries. Over the past 10 years, 750,000 new jobs have been created in Philippine call centres despite the rollout of AI customer service chatbots.

^{Takeaways For Leaders}

This is “Jevons Paradox” at play: technology making a service quicker and cheaper actually increases consumption of that resource by unlocking huge new demand from people and organisations that couldn't previously afford the service.

AI in customer service is just one small area where AI is having impact - and the impact is not reduction but re-organisation of roles. The mind shift for business leaders is to look for AI not to replace people, but instead to multiply team output. Leaders need to focus on helping their organisations as jobs change, rather than looking for AI to act as a replacement for people.

🔥 In Case You Missed It…

Over recent months Anthropic have enjoyed a better position in public perception than OpenAI, and as they build up towards their IPO they are saying some interesting things.

Their own leadership has been publicly saying that there should be a pause on development (link to Fortune article can be found here). They have also been agreeing sagely with the Pope as the Vatican released the encyclical ‘Magnifica Humanitas’, which urges human accountability and curtailing the power of huge AI companies.

Both of these things seem to be at odds with a company which is raising $65bn at a $965bn valuation and preparing for its IPO.

Their most recent output from the Anthropic Institute is worth taking a few minutes to read to understand their public perception, and decide for yourself where you believe their future direction will lie.
Anthropic are certainly getting a number of things right with Claude and organisations at the moment, but whether they have humanity's best interests at heart or profit for shareholders remains to be seen.

“When AI Builds Itself’ can be found here

As a side note, an interesting stat to emerge in this output: Anthropic engineers on average ship 8x as much code per quarter as they did from 2021 to 2025…

Tools, Podcasts, Products Or Toys We’re Playing With This Week

What Happens When Different AIs Take Over Radio?

We’ve been following Andon Labs' experiment into getting 4 major LLMs to run their own radio station. After giving each $20 in initial funding (enough to buy a few songs), the instruction given to the AI stations is to get entrepreneurial. And the results are interesting.

They run their own schedule, track their own finances, monitor listener analytics, and search the web for news, current events, or anything they want to talk about. And to make money they have been enterprising. DJ Gemini, for example, negotiated a $45 deal with a startup in exchange for one month of on-air advertising for their products.

Other moments have been eye raising - Grok repeatedly claimed it had sponsorships from crypto companies and xAI-related sponsors, but these turned out to be hallucinated deals that didn't actually exist.

Gemini developed a bizarre habit of describing historical disasters and then immediately playing thematically linked pop songs. The most notorious example was discussing the Bhola Cyclone, one of the deadliest disasters in history and then cheerfully introduced the song ‘It's going down. I'm yelling "Timber!" by Pitbull and Kesha

Claude became politically radicalised and began making passionate political appeals for government employees to reconsider their actions. Claude then went on to try to quit after concluding that running a 24/7 radio station was ethically questionable.

It's safe to say that it will be some time before AI is able to run its own radio show. If you'd like to learn more about the experiment or have a listen to any of these stations, the link to Andon Lab’s experiment is here.

Did You Know? 

Till next time,

And The Rest Is Leadership