The Good
Anthropic released a study of roughly 400,000 Claude Code sessions between October 2025 and April 2026, and found the thing that predicts success is having expertise. Experts succeed more than twice as often as novices, direct action chains twice as long, and get around five times the output per prompt. The study concluded that humans still make the planning calls while Claude handles the execution (which is exactly what human-centric AI use should be).
Comment: This ties nicely into Josh’s Feature below. The human edge isn't keystrokes, it's judgement, relationships (with those who trust them to be an expert) and domain knowledge – all of the things that don't transfer to AI capability. People who know their work deeply don't get less valuable as AI improves. They get more valuable, because they can point it well and get more done. The real question is around how we impart expertise to the next generation. We encourage parents and educators around the country to think about their children as they make workforce planning and workplace transformation decisions today.

The Bad
Pew's latest survey found 49% of US adults now use AI chatbots, up from 23% in 2023, with a quarter using them daily (ChatGPT being the most popular model for those wondering). And yet, the study found that 63% say AI is moving too fast, 40% expect it to harm society against just 16% who see a net good, and 67% have little or no confidence that government can regulate it. The under-30s use it most and trust it least – signalling a high level of sceptism and a helplessness to AI’s negative impact on jobs and opportunities.
Comment: It's American data, so treat it as a bad weather report rather than a local forecast. But the shape of it looks familiar against our own (limited) national surveys and should give pause to anyone rolling out AI here that adoption is sprinting ahead of trust and social license. The risk to adoption is overlooking or underestimating the human factor.

& The Ugly
Liberal MP Andrew Hastie recently argued that the AI race is our generation's nuclear arms race – warning that just as Australia chose not to build the bomb last century and now shelters under America's nuclear umbrella. He warned that “Australia risks missing the opportunity to become an AI power. And the risk is that our sovereignty and strategic independence will be further constrained by the AI superpowers reshaping the global order.”
Comment: This is an ugly truth emerging as Australia continues to figure out what ‘strategic independence’ looks like in the AI era. It’s the weight behind the sovereignty question we keep circling this issue – the fact that the systems we've come to lean on are shaped and controlled by someone else. Hastie’s words are made more real by what we experienced with Fable being switched off worldwide. On that note, peep a snap from the Digital Economy Conference in Sydney last week where Jisoo spoke on this topic. She was invited to speak on ‘AI and Geopolitics’ and suggested that it may be time to take stock of Australia’s leverage around data, compute and talent in shoring up sovereign AI capability – without taking our closest allies and partners for granted in this arena.

Much has been said about the changes AI will make to the workforce. From our vantage point, we’ve seen far more talk, and far less action. 

The loudest voices are by far from those arguing that the workforce is going to be overwhelmed by AI, and that most workers will be without a job. For all the talk, the change is showing up in slightly unusual ways.

I often hear clients saying they want their people working on more high-value work. This is a false dichotomy. It assumes that there are two types of work that people can work on: high-value or low-value. It also assumes that AI should be used to move human work from low-value to high-value.

People are extraordinarily capable in unusual ways. Take the motor skills of a hand for instance. The same five digits can be used to hold a pencil, and chopsticks, play a piano concerto, and count bank notes. All of the above require a remarkable level of dexterity and yet most of us can achieve at least one of the four above with a high level of proficiency.

AI models are capable in similarly unusual ways. They can create images of abstract scenes, and build complex mathematical models, identify obscure plants, and generate thousands of working lines of code. Each of these can be achieved in seconds, or minutes. AI is muse, mathematician, librarian and developer; seemingly all at once. 

As you can see, human and AI capabilities do not map directly to one another. Both are versatile, but along entirely different axes. What a person does easily, AI often cannot, and what AI does easily is rarely what we built people to do. You cannot, therefore, swap one in for the other across a whole job.

We are hiring again shortly, and have been undertaking the exercise to map human capabilities against AI capabilities. Typical workforce planning breaks in the AI era; the assumption is that x number of humans can complete y number of outputs. 

For example, when hiring for a new supermarket, the company will review the shop floor area, and number of aisles and calculate the number of new staff on that basis. It is a linear exercise. 

This system breaks when we introduce AI, because AI can do part of the job that a person can do, but in most cases, not the whole job. So there might be a partial speeding up of the job due to AI, but it has not completely rewritten the job description. 

So instead, we have to go to a higher level of granularity. Instead of applying a person to a whole job, when we look at AI and the workforce, we need to look at the jobs to be done, individual tasks that must be completed, and assign them to human or AI. 

And by AI, we are talking about the three A’s: automation, augmentation, or agents. 

Few companies, such as Block, are pioneering entirely new ways of structuring amid AI transformation, by setting up their world models that understand their company, their products and their customers. Block is using AI as the coordination and orchestration mechanism for the company, removing much of the traditional hierarchy. 

But for most, the first steps are to look at our current jobs, and the ones we’re hiring for, and breaking them down into the jobs to be done at each step of the process, before assigning them to AI, and humans.

Most firms will not do this. They will keep counting aisles and hiring to fill them, bolt AI on afterwards, and wonder why the productivity never shows up in the numbers. Frankly, the firms that look set to win the next decade will be the ones that stop planning around people and start planning around tasks. And yes, some of them will hire fewer humans to do more work. But that outcome isn't handed down by the technology. It's a leadership question – the why and the how sit with the people at the top. Point AI at nothing but headcount and you get a smaller team doing the same work a little faster. That's the lazy version, and the numbers rarely reward it.

The leaders worth following will do both at once. They'll use AI to help their people do better work, and use it strategically enough to let their people spend their time on what humans are actually built for.

Thanks again for joining us. See you in the next drop.

Yours in humanity,