Schrödinger’s Optimism
Reading news stories about the US stock market dip at the end of last week, you might think that serious economic and technology analysts are uncertain about the impact of AI on business and productivity.
Selling or buying stocks is quite a binary activity (notwithstanding the grey areas of hedging and options), but the current state of AI is more quantum than binary – simultaneously beyond imagination and yet not good enough for deployment in production; able to autonomously code entire apps with a few lines of instruction, but struggling with basic maths or questions like “should I drive to the car wash?”
We will probably need to live with its patchy, jagged, probabilistic, non-evenly-distributed nature for some time, and co-evolve our methods with the technology in much the same way as quantum computing relies on error correction. The question now is how quickly our institutions can metabolise what the technology is already making possible.
But if you zoom out just a little, the progress being made is incredible, and would have been unimaginable just a few years ago.
Matt Shumer recently wrote a widely-shared article trying to put this progress into words, which started with the reflection that people who are not following AI developments don’t know quite how disruptive the next 5 years could be:
I’ve spent six years building an AI startup and investing in the space. I live in this world. And I’m writing this for the people in my life who don’t… my family, my friends, the people I care about who keep asking me “so what’s the deal with AI?” and getting an answer that doesn’t do justice to what’s actually happening. I keep giving them the polite version. The cocktail-party version. Because the honest version sounds like I’ve lost my mind. And for a while, I told myself that was a good enough reason to keep what’s truly happening to myself. But the gap between what I’ve been saying and what is actually happening has gotten far too big. The people I care about deserve to hear what is coming, even if it sounds crazy.
He goes on to discuss what this means for jobs, extrapolating from his own experience as a software developer working in AI, and it is simultaneously an exciting and also very discomforting read. But it is also naive.
On the optimistic side, I believe our lives should not be dictated by “jobs”, which have become hollowed out and insufficiently remunerated to live well at the entry level to mid-tier (at least outside of tech). But realistically, in the absence of labour market improvement or Universal Basic Income (UBI) or any other idea about how young people without assets can support themselves, the implications of what Shumer predicts could be very worrying.
However, business and societal change is modulated by incredible reserves of inertia that can hold back progress for decades, if not centuries, as long as enough Powerpoint enjoyers leaders are invested in the old ways of doing things.
A case in point is the debate about SaaS platforms in business:
- Logically, many of them are screwed.
- Practically, firms can now recreate better, simpler versions of them without the eye-watering subscription costs using coding agents.
- Emotionally, they weigh so heavily on employee experience that companies would be far happier if they ceased to exist.
And yet … don’t count them out. There are some very human – and very illogical – reasons on both the buyer side and the vendor side that suggest these businesses might not be so easy to kill, as Finbarr Taylor argues here. Just because a better way is possible, it doesn’t mean it will come to pass:
You don’t always pick the cheapest option. You don’t always pick the most innovative option. You pick the option that, if it fails, you can defend to your boss. “We went with Salesforce” is a defensible sentence in any boardroom in America. “We went with an app I vibe-coded over the weekend” is a resignation letter.
This is the same dynamic that kept IBM dominant for decades and that keeps McKinsey and Deloitte in business despite armies of cheaper, often smarter competitors. Enterprise buyers optimize for career risk, not unit cost. They want a vendor that will still exist in three years, that has a support team they can call at 2am, that has a track record of not losing their data.
Change is not inevitable. At least not everywhere.
Exponential Proponents
This weekend, Azeem Azhar also published an eye-popping piece about the speed of AI’s evolution, predicated on the realisation that he had consumed 97 million tokens in a single day of working with AI tools. He makes the point that in exponential change, each level of scale can be fundamentally different from those below them. At 10^3 tokens, AI is a toy. But as we add a zero to the tokens used, it becomes a tool, then a colleague, a workflow, a process and a workforce; but at 10^9 tokens, it is more like infrastructure – always on, always working, like electricity.
At 10⁹, a billion tokens a day per person, our unit of analysis changes. This becomes agents spawning even more sub-agents and talking to them and other agents. The human sets direction and adjudicates edge cases, but the conversation is mostly not ours anymore.
I’ve already caught a glimpse with my own setup. Micromanaging them slows down the whole process. If you had to configure each sub-agent yourself and track their work, I’m pretty certain none of us would do it. In other words, the bottleneck is no longer the model’s capability; it’s your willingness to let go. It becomes like running an organisation, trusting the parts to make the whole.
And for those at the frontier of AI usage, the tools are not reducing effort, or giving them extra leisure time; in fact they are intensifying their work, as Aruna Ranganathan and Xingqi Maggie Ye found in their recently published eight-month HBR study of 200 workers at a hi-tech firm, which raises some interesting questions for leaders:
The promise of generative AI lies not only in what it can do for work, but in how thoughtfully it is integrated into the daily rhythm. Our findings suggest that without intention, AI makes it easier to do more—but harder to stop. An AI practice offers a counterbalance: a way to preserve moments for recovery and reflection even as work accelerates. The question facing organizations is not whether AI will change work, but whether they will actively shape that change—or let it quietly shape them.
Which way up is that J-curve?
In the FT this weekend, Erik Brynjolfsson made the case that AI-attributed productivity improvements are starting to show up in the data (also commented on by Andrew McAfee here if the FT link is paywalled):
Data released this week offers a striking corrective to the narrative that AI has yet to have an impact on the US economy as a whole. While initial reports suggested a year of steady labour expansion in the US, the new figures reveal that total payroll growth was revised downward by approximately 403,000 jobs. Crucially, this downward revision occurred while real GDP remained robust, including a 3.7 per cent growth rate in the fourth quarter. This decoupling — maintaining high output with significantly lower labour input — is the hallmark of productivity growth.
My own updated analysis suggests a US productivity increase of roughly 2.7 per cent for 2025. This is a near doubling from the sluggish 1.4 per cent annual average that characterised the past decade.
Does this represent the beginning of the hoped-for productivity J-curve promised by AI optimists? Or are we seeing business leaders using automation to shed jobs, whilst protecting their own, with no reduction in overall output? Or, is the mild increase in US GDP nothing to do with technology at all, and could negative payroll growth indicate recessionary dynamics down the line? We will see.
It is sad to see leaders of large, established organisations respond to abundant technological capability by cutting the junior headcount (a.k.a their future) just to appease the fickle stock-trading gods in a time of market volatility, or to protect themselves until they can exit. If ever there was a time for long-term thinking about organisational development, it is now. Maybe private companies and those owned by long-term family trusts will be among those to chart a path through this fear and end up as winners.
But for individuals trying to get by in this liminal space between old and new worlds, it could be challenging. The most empowered, high-agency individuals and teams can achieve more than ever, but many of the entry-level jobs young people have been conveyed towards since they started school may not exist (or at least in such numbers) in the near future. If you have agents to manage, you might make it. But outside tech, old-fashioned management structures demand an awful lot of pointless busy work at the base of the pyramid that might start to be replaced sooner than we think.
We need to focus on helping the best leaders move closer to the work, not retreat further into abstraction and politics. AI makes it possible to compress layers, to give experienced people direct leverage over real outcomes rather than managing proxies and reports. But that only happens if incentives shift. In many firms, status is still measured by distance from execution, and risk is minimised by preserving familiar structures. Unless those dynamics change, AI will be used to thin out the base of the pyramid while leaving its shape intact. That would generate marginal gains at best, but reduce the organisation’s capacity to explore and exploit the kind of exponential gains that Matt Shumer and Azeem Azhar believe are possible.