Is The AI Future Of Work Really That Dark & Dystopian?
I’m proud to announce that the next cohort of my Data & AI Strategist Certification is open for enrollment. It begins on June 3rd and runs for six weeks. Learn more and enroll here.
Sorry for the duplicate email for paid subscribers. This was intended for everyone, but was only sent to paid subscribers.
What’s next for workers is hotly debated, but most opinions are skewed by AI hype and a very myopic view. Take Klarna, for instance. Some see Klarna’s decision to aggressively adopt AI and replace workers as a sign of things to come. It reduced headcount by almost 40% over 2 years.
Now, the company plans to rehire some customer service personnel because the quality of work suffered when they transitioned to 100% AI. Klarna’s CEO wants to ensure that customers always have the option of talking to a person. So, which is it: AI replaces employees, or AI doesn’t work?
The duality is a common theme. Early adopters of the ‘AI replacing employees’ cycle are having some buyer’s remorse. You could look at Klarna and see proof that AI cannot replace people. You could also say that the company is unlikely to rehire all the people who were let go. Even if the final staffing reductions due to AI are only 20% or 30%, it’s still significant.
According to a recent Georgia State University study, which evaluated the impacts of AI automation, firms saw lower labor costs when they replaced workers with AI, but not overall productivity improvements. Essentially, replacing people with AI is more cost-effective, but the AI itself isn’t more productive than the people it replaces.
Efficiency increases only occur when businesses augment people with AI. The future of work isn’t replacement (yet) or AI failure. As with most technology cycles, the implications are more nuanced and the impacts more complex. As good as the augmentation paradigm sounds, it won’t prevent job losses or significant workforce disruptions.
The shift to our AI-augmented future of work is already underway. Why haven’t you heard more about it? Look no further than Duolingo. The backlash to AI has been significant. Put simply, businesses are downplaying the impact of AI on their workforce because it’s hugely unpopular.
Coverups I Am Seeing
I know many companies that are actively working to conceal the impact of AI on their workforces. These findings come from the competitive intelligence V Squared does for our clients.
Many companies that recently announced layoffs targeting middle managers and low performers are hiding AI augmentation impacts in those numbers. Microsoft’s recent layoffs are a good example. The company wants to frame this round as targeting middle managers, but the numbers show that software engineering was the biggest category of impacted workers, and many of them were not in leadership positions.
Others are framing layoffs as a process of slimming down and right-sizing to build a more efficient business. It’s an old line that’s being dusted off to obscure the jobs that are being lost to AI.
Some businesses are simply implementing a hiring freeze. Normal attrition rates run between 10% and 20% per year. Companies are laying people off by waiting for them to leave on their own and not replacing them.
US and EU-based companies are offshoring teams while they experiment with AI augmentation or outright replacement. They don’t have to advertise layoffs due to AI if they impact an outsourced team, cut contractors, or reduce their reliance on external service vendors. Cognizant, Infosys, and TCS are ramping up to handle a surge in demand. I fear that there’s a quick reversal coming in the next 2 years as AI augmentation and replacement initiatives come to fruition.
The Cycle Is Picking Up Steam
Drugstore chains are struggling to remain profitable. Customers aren’t buying the higher-margin convenience store-type items that had helped drive profits in the past. They’re also dissatisfied with the experience and are turning to other stores, such as Costco and Walmart, to fill their prescriptions. Even Amazon is attempting to elbow in on their business.
Walgreens is turning to automation to improve the customer experience. It doesn’t have the margins to hire enough pharmacy workers, so robotics is its long-term solution to improve service levels. It has built micro-fulfillment centers where robots fill 40% of prescriptions for the pharmacies they support. Walgreens has plans to increase the number of pharmacies served by these micro-fulfillment centers to 5,000 by the end of the year.
It's a common response to rising competition. While Klarna and Duolingo are turning to AI to help meet investors’ aggressive growth expectations, traditional companies have other reasons to embrace automation. Walmart increased its reliance on robotic automation to be more competitive against Amazon and Target. The restaurant industry is broadly evaluating robotic automation for its kitchens and service to offset declining consumer demand and rising costs.
What Does The AI Augmentation Paradigm Look Like?
In my book and courses, I teach a maturity model framework for AI augmentation with five phases, but only three are relevant for this article:
Human-Machine Tooling
Human-Machine Teaming
Human-Machine Collaboration
Digital software and simple machine learning models follow the first phase, where people use them as tools, like a hammer. They make the job easier, so more people are capable of doing it, and make people more efficient. More complex automation follows the second phase, where people supervise agents or robots while they do work.
How close are we to that time? Amazon released Vulcan last week, a new line of robots that work in its warehouses. The top-line announcement was focused on the robots’ sense of touch and how it enabled them to do their jobs more safely. It started the conversation on human-robot augmentation with a focus on how Vulcan will prevent people from having to bend or reach above their heads and reduce stress injuries.
Separately, Amazon explained that it would be training a small number of warehouse workers as robot technicians and supervisors. Its description fits with my Human-Machine Teaming phase. Initially, people and robots work side by side, as they do in Amazon’s warehouses today. Over time, people do less and less work as robots become capable of reliably managing a greater share of the workflow. That shifts people from doer roles to monitor, supervisor, and quality assurance roles.
One person can oversee 7-10 robots today, and that number will increase as they become more reliable. If a robot is as productive as a person, we can assume that Amazon’s need for people will drop by 70%-80%. New roles will be created for robotic site reliability and maintenance engineers, but only a small number of warehouse workers are being reskilled. New roles won’t offset the job losses.
The construction industry is using robotics following the Human-Machine Teaming phase as well. The pictures from this article illustrate teaming in the real world. A crew of 5 was able to accomplish what typically requires 25-30 people. Walmart and Starbucks are prominent early adopters of robotic assembly. However, construction robotics is approximately 2-3 years behind the capabilities of warehouse and restaurant kitchen robotics. Still, the paradigm is consistent across domains.
What’s The Worst That Can Happen?
This article is easily the darkest take on the future of the technology field that I have ever read. It predicts a time when Amazon’s approach to warehouse worker and truck driver efficiency comes to technical workers. Software development metrics are tracked down to lines of code per minute and defects per line. Any variation from prescribed efficiency metrics results in disciplinary write-ups and eventual termination.
The author argues that technical roles will be reduced to prompting, validating, and fixing the output of AI agents. They will be algorithmically managed and held to increasingly unrealistic productivity and quality standards. The author believes that the only thing stopping major tech companies from doing this today is the current scarcity of digital and AI knowledge workers. He suggests that tech companies are laying off workers not for profitability, but to accelerate the reversal of this power dynamic.
I don’t shy away from a dark take on corporate culture and the future of work, but the author leans into the worst-case scenario. It’s tempting to dismiss him, but there’s evidence to support his case. His article focuses on Amazon, but algorithmic monitoring and workforce maximization are prevalent.
Maximization is a rising star in the corporate efficiency and business model optimization consulting world. Workforce maximization is the idea that we can define the perfect employee by a set of behaviors that maximize their contributions to the business. Operational excellence, lean operations, and talent density are part of that push.
Consultants use models and digital twins to algorithmically optimize workflows and define the ideal process. Employees are retrained and held to the algorithmic standard. The challenges have consistently been enforcing compliance with the new standard and the time it takes employees to adopt the new workflow. Consultants would like to make more frequent changes and run experiments, but every optimization comes with significant costs and disruptions.
AI agents can manage frequent workflow changes with near 100% adoption and little disruption to operations, quality, or service levels. They are more willing to be the perfect employee and operate as instructed without deviating from the optimal workflow. How likely is it that the changes happening in physical roles will spill over into software engineering and other non-physical roles?
From Robots To Agents
Replace robots with AI agents, and the Human-Machine Teaming phase applies just as well to non-physical roles, such as software engineering, customer service, and marketing. Software engineers supervising a team of 7-10 agents fit the paradigm, and AI coding tools are improving quickly.
Meta and Pinterest are scaling AI automation for marketing, and the same thing could happen in that domain. SAP is deploying agents and plans to support hundreds of operational use cases. Salesforce is scaling agents for customer service, finance, and sales.
If you’ll be in NYC on May 21st, you can meet me and see the latest Agentforce updates at Salesforce’s next World Tour event. It’s free to attend, and you can register here.
Klarna is transitioning its customer service teams back to a Human-Machine Teaming phase after initially jumping straight to replacement and realizing the technology isn’t ready for that, yet. One customer service agent will supervise multiple AI agents and step in when customers need human intervention.
Salesforce has leaned into the AI augmentation paradigm, with Marc Benioff encouraging customers to take an abundance mindset rather than embracing the dystopian AI replacement paradigm. In his view, AI will create growth opportunities, and the business will need to scale to take advantage of them. AI agents offer a path to scaling that offsets much of the new hiring with AI efficiency and productivity gains. Businesses will scale faster, and margins will rise.
If this sounds like my Core-Rim framework, you’re right. I don’t know if Marc Benioff or Sebastian Siemiatkowski has read my book, but people from Klarna and Salesforce have taken my courses. A growing number of businesses are adopting and delivering on AI strategies that align with my frameworks, so who knows? I have some amazing people in my certification network, and they may be gaining traction.
Business operational workflows have different levels of complexity. Technology can manage some types of complexity, and most operating models now run on a technology core. Some complexity cannot be managed by technology, and people manage the rim of the operating model. Technology reduces the business’s complexity, making it more transparent, easier to manage, and quicker to transform. Benioff and many other executives see that as the blueprint for modern AI-augmented businesses.
The extent to which each domain advances its technology core and Human-Machine Teaming phase depends on one thing…
The Unreliable Elephant In The Room
AI augmentation is limited by one factor more than any other: reliability. In my maturity model, each phase requires a significant jump in model reliability to gain user trust and drive adoption. Agents will only reach the Human-Machine Teaming phase if they require significantly fewer people to supervise and maintain them than it does to manage the workflows manually.
Once we give an AI agent or robot the agency to act in the real world, it must be reliable enough to do the job well. Robots can’t go rogue and injure people or deliver products that require constant repair. Agents can’t color outside the lines and delete the code base, or make updates that crash production.
Robots and agents improve by doing and getting feedback from people. My reliability framework states that every time an agent’s usage doubles, its reliability gap (the difference between human expert and model performance) is cut in half.
We’re seeing growing real-world AI adoption and usage, so models will improve through multiple avenues of reinforcement learning. Unless the model imposes a fundamental limitation, data and usage are the only constraints on reliability. As the dataset (built through observation or feedback) gets more complete and becomes a knowledge graph, the robot or agent will become more reliable. Practice really does make perfect.
Learning by doing leads to compounding improvements in my experience. Models get better in leaps, and those leaps become more frequent as more users join the platform. The more cases and variations the model sees, the faster it improves. The adoption curve makes me believe that coding, marketing, sales, finance, HR, and customer service agents will improve significantly in the next 2 years.
The more an agent or robots gets used, the faster it achieves the reliability necessary for Human-Machine Teaming. That will change the nature of the roles the agent or robot augments and reduce demand significantly.
How Long Until AI Replaces People?
As agents become more reliable and people trust them more, we will enter phase three, Human-Machine Collaboration. People will give agents true agency to complete tasks and achieve outcomes. AI replacing employees becomes feasible at this phase.
That will quickly lead to Agent-Machine Teaming and Agent-Agent Collaboration. However, in nearly every case, we’re still 3+ years away from this. We’re only seeing agents and robots take over small parts of our workflows. Handing a complete workflow over requires a much higher reliability level than our current models possess.
What’s more, only a few companies have figured out AI product, platform, and agent design patterns. The models are just one piece of a much more complex agentic system and product puzzle. Even when companies can meet use case reliability requirements, most will struggle for a year or more with productization and commercialization.
Adoption presents additional challenges. These platforms don’t simply bolt on to the business; process engineering, knowledge engineering, and transformation are required. Few enterprises possess those capabilities, and without them, AI platforms and agentic systems won’t deliver much value. Vendors must transform into partners who help businesses do all the work necessary to adopt and realize returns.
What Should You Do to Prepare?
Most advice focuses on becoming proficient in using AI in your current role. That’s shortsighted. If you understand the Human-Machine Teaming paradigm, you see where that leads you: decreasing job prospects in a low-value role. The last place you want to end up is as an agent supervisor or quality control tech.
For a couple of years, these roles will appear very appealing. Hyper-productive software engineers, salespeople, and marketers will be held up as the future of work. However, that will quickly change as agents become more reliable and the skill level required to supervise an agent team decreases. Take advantage of the trend, but prepare to transition into another job category.
In my Core-Rim framework, the employees who manage irreducible complexity fall into three categories.
Laborers Follow Processes & Deliver Artifacts
Knowledge Workers Use Frameworks & Deliver Outcomes
Strategists Implement Frameworks & Control Transformations
Agent supervisors and quality control techs fall into the first category. Physical and digital laborers will be hardest hit by agents’ increasing reliability. Some laborers will be upskilled into the knowledge worker category. As jobs become outcomes-focused, a new layer is emerging in the business. Knowledge work is no longer siloed to a single domain.
Software engineers need product, engineering, and business domain expertise to transition from laborers to knowledge workers. That puts everything required to understand the business need, develop the product, and deliver it to laborers in the hands of a single role. Knowledge workers are capable of delivering the business outcome rather than just the technical artifact, and they can do it with minimal support.
We’re already seeing job descriptions and roles transform in this direction. Over the next 2 years, demand for technical labor will fall while technical knowledge workers will be a growth area. The trend will extend beyond just engineering domains. For example, marketers will require engineering and product capabilities to meet more of their own technical needs. The cycle has begun with job descriptions expanding to include the use of AI tools and self-service data or development platforms.
Every role will gain multiple facets, although not all will follow the same structure. As I often say, what it means to be technical is changing. Coding won’t be the only high-value technical capability. Building knowledge systems is another that’s rising fast, and more are coming soon.
Business and operating model engineering is another technical role. This is where the strategist category takes on new technical capabilities. Executives in a high-maturity business no longer manage tasks, workflows, or work products. They implement frameworks and control transformations to deliver new, increasingly valuable business and customer outcomes.
This may all seem far off, but most of these transformations are happening now. We’re seeing the early signs of a much larger shift. As with every major technology disruption, jobs will change gradually at first, then undergo more significant disruptions.
Dramatic change looks dystopian and dark because it’s often difficult to see where we fit in the new order. Many of us will transition and upskill into new job categories. The sooner we begin adapting, the better the future looks.
At the same time, I don’t want to gloss over the people who will fall behind. Many will refuse to adapt until they’ve been laid off and are unable to find a new job. Others will be left behind because they didn’t see what was coming or didn’t have the option to upskill. The future of work isn’t that dystopian, but that’s not to say everyone will benefit. If we’re not intentional about how we bring workers forward, a large portion of the workforce will be hurt by AI augmentation.