An AI expert claimed his agent replaced 1000 people; the reality is more interesting and more useful to leaders weighing AI investment.
I heard an apparently reputable “AI expert” the other day say that his AI agent had “replaced the work of 1000 people”. I was at first sceptical, then intrigued. Then I asked myself a simple question: which 1000 people?
Is it 1000 fruit pickers? The 1000 footballers of a FIFA World Cup? The massed chorus and orchestra of Mahler’s Symphony No. 8, the “Symphony of a Thousand”? I suspect it was none of these. In fact, I suspect that this gentleman’s army of bots did not replace 1000 people at all.
Doing the work of 1000 people with a computer is trivially easy. I could write a script to print the letter “A” the same number of times as 1000 people could write it in a day. Trivial, but worthless. Doing valuable work already performed by 1000 people, to a higher standard, more quickly, or more cheaply, and then actually replacing those 1000 people along with their office space, machinery, and pension entitlements: that’s entirely different. I suspect that’s why the “work” of 1000 people was claimed, rather than the useful jobs of 1000 people.
Here’s what I think: AI agents will transform how we work, but they won’t eliminate jobs at scale. They’ll compress tedious tasks, unlock previously uneconomical work, and create entirely new categories of valuable activity. The organisations that thrive will be those that understand this distinction.
What AI agents actually do
An AI agent differs from a chatbot or simple automation in three critical ways. First, it can break down complex requests into steps, execute them, and handle obstacles without constant supervision. Second, agents interact with various tools and services: accessing databases, calling APIs, running code, manipulating files, or controlling software. Third, rather than giving a single response, agents can try different approaches, evaluate whether they’re making progress, and adjust their strategy.
This sounds revolutionary. And in many ways, it is. But revolution doesn’t mean replacement.
The Klarna lesson: When 700 becomes zero, then returns
Consider Klarna, the Swedish fintech firm that provides perhaps the most instructive case study in AI agent deployment. The company initially replaced much of its customer service staff with AI chatbots, claiming they performed the work of 700 employees. Note: not quite 1000, but close enough for our purposes.
The result? Service quality declined. Customer satisfaction plummeted. Within months, Klarna began rehiring human agents. CEO Sebastian Siemiatkowski acknowledged that their over-reliance on cost-cutting had led to poorer service, and emphasised instead the necessity of human interaction for customer satisfaction. The company now employs remote, contract-based human agents, having learned that complex, empathetic customer interactions still require human judgement.
This isn’t a story of AI failure. Klarna’s chatbots, I’m sure, handle routine queries perfectly well. It’s a story about understanding boundaries: where efficiency gains become service degradation, and conversely, where automation enhances rather than replaces human capability.
IBM’s HR automation
IBM provides an example of a more successful approach. They used AI agents to automate hundreds of back-office HR roles: benefits processing, employment verification, standard documentation, and answering routine employee queries. As CEO Arvind Krishna noted in a Wall Street Journal interview, this enabled expansion, not reduction: “While we have done a huge amount of work on leveraging AI and automation on certain enterprise workflows, our total employment has actually gone up, because what it does is it gives you more investment to put into other areas”.
The resulting investment in this case went into sales, marketing and software engineering, and HR professionals shifted from processing to strategic work. The company grew; the roles evolved. Notably, the work of around 200 people was replaced, amid a more complex backdrop of reorganisation. Krishna emphasises that the work that was unlocked involved critical thinking, where people “face up against other humans, as opposed to just doing rote process work”. In HR at IBM, therefore, AI is not managing conflicts or handling sensitive personal employee issues – this is what the humans are good at, and they now have more time to do it well.
This transformation requires precision, and IBM targeted purely transactional tasks. By comparison, Klarna tried to automate relationship-based work and failed. There are clear learnings about the boundaries involved: automate transactions, augment judgement, amplify human relationships.
The Paralegal paradox
Let’s explore another example: paralegal work. Reviewing contracts, identifying relevant clauses, flagging risks, extracting key terms, checking compliance against standards. This work requires legal training and meticulous attention to detail, yet it’s ideal for AI agents. Document automation platforms and eDiscovery tools already handle much of this work faster and more accurately than humans.
So why are there still 1000 paralegal vacancies in the UK (on LinkedIn) at the time of writing?
The answer reveals something crucial about technological adoption in professional services, an echo of what I’ve described at IBM. Law firms are buying these tools for efficiency gains over headcount reduction. Paralegals now process more documents, handle more complex cases, and deliver higher-value analysis. The bottleneck hasn’t been eliminated (it never is), but it’s been shifted. The paralegal who once spent days reviewing standard contracts can now spend time on nuanced regulatory compliance issues that AI cannot – yet – navigate.
This pattern repeats across professional services. AI agents excel at high-volume, rule-based tasks. They struggle with ambiguity, novel situations, and anything requiring genuine human judgement about human matters. More importantly, they cannot bear legal responsibility, build trust with anxious clients, or make high-stakes decisions where catastrophic failure is possible.
A real opportunity: unlocking latent demand
Every efficiency gain creates new possibilities. This is what the “1000 people replaced” narrative misses. When spreadsheet software arrived, it didn’t eliminate accountants; it enabled financial modelling that was previously impossible. When CAD replaced drafting tables, it didn’t eliminate architects; it enabled iterative design processes that transformed the built environment.
This effect even has a name: the Jevons Paradox. In 1865, the English economist William Jevons noticed that improvements (from technology) that increased the efficiency of coal use actually led to an increase in the consumption of coal. Contrary to intuition, he argued that progress in technology could not be relied upon to reduce fuel consumption. This point, in the field of energy consumption, has of course been proven many times over since 1865.
I’m seeing that AI agents follow this pattern. They’re best deployed not to eliminate roles but to unlock work that’s currently:
- Too expensive to justify (comprehensive contract review for small businesses)
- Too time-consuming to attempt (personalised customer communications at scale)
- Too complex to coordinate (multi-system data reconciliation)
- Too tedious to sustain (continuous compliance monitoring)
Organisations that understand this deploy AI agents strategically. They identify bottlenecks where human expertise is wasted on routine work. They automate the mechanical to amplify the creative. They use AI to do more, not to do the same with less.
What this means for you
The question isn’t whether AI agents can do the work of 1000 people. It’s whether your organisation can identify the right work for AI agents to do. This requires:
- Process maturity: You can’t automate chaos. Successful AI agent deployment requires well-defined processes, clear success metrics, and robust feedback loops.
- Strategic clarity: Which bottlenecks actually matter? Where would efficiency gains translate to competitive advantage rather than marginal cost savings?
- Implementation expertise: The gap between AI capability and AI deployment is vast. It requires technical architecture, change management, risk assessment, and continuous optimisation.
- Realistic expectations: AI agents augment; they don’t replace. The organisations seeing real returns understand this and design accordingly.
At Softwire, we’ve guided dozens of organisations through challenging change and transformation. We’ve studied the failures (like Klarna’s initial overreach) and the successes (systems that genuinely unlock new value). We know which processes benefit from AI agents and which require human judgement. Most importantly, we understand that successful AI deployment isn’t about replacing your thousand people; it’s about enabling them to deliver greatly more value.
The “AI expert” claiming to replace 1000 people is, I’d say, selling a fantasy. The reality is both more modest and yet more exciting: AI agents can transform what your existing team can achieve. They can eliminate drudgery, accelerate delivery, and enable work you couldn’t previously contemplate.
If you’re ready to explore what AI agents could genuinely do for your organisation, without the hyperbole and with a clear view of both opportunities and boundaries, I’d welcome the conversation. Because while I won’t offer to replace the work of 1000 people, I can show you how to make your current team significantly more effective. In today’s market, that’s worth far more than empty promises about replacement.
