Key Takeaways
Redesign whole workflows around human-machine teams, not isolated automation.
The real opportunity is what the authors call the missing middle: work redesigned so people and AI systems collaborate, each contributing strengths the other lacks. Machines can sift huge datasets and handle repetition; people supply judgment, context, and the ability to navigate ambiguity.
The authors describe three shifts in business: standardized processes, automated processes, and now adaptive processes that respond to real-time information. Their navigation analogy makes the change vivid. A basic GPS follows a preset route; Waze updates directions using live traffic reports and reroutes as conditions change. Likewise, simply automating an old workflow may improve efficiency, but redesigning it around people, machines, and current information can change what the process can achieve.
What's striking is the shift in the unit of analysis: the question is not whether a job survives, but how the work itself should be organized. That aligns with sociotechnical design, which treats tools, roles, incentives, and communication as one system. A new tool inside an unchanged workflow often just speeds up the old bottleneck. Still, collaboration is not automatically superior. It requires clear handoffs, meaningful human authority, and enough training to make the partnership real rather than a machine recommendation that employees are expected to rubber-stamp.
Let machines absorb repetition so people can handle ambiguity and judgment.
Divide work by strengths. In a global bank, machine-learning tools for detecting possible money laundering reduced false alerts by as much as 30 percent and investigation costs by 40 percent. Staff could spend more time on cases where financial patterns needed human interpretation and compliance expertise.
Virgin Trains redesigned its complaint process in a similar way. Software reads, sorts, and routes customer messages, then prepares case files for employees. It handles routine complaints and flags uncertain cases for human review. Manual work fell by 85 percent, while the company increased correspondence by 20 percent by accepting complaints through more channels. These examples show the difference between replacing a worker and removing the dullest steps from a worker's day. The saved time matters only if organizations use it for more valuable work.
This is a practical version of comparative advantage: assign each task to the party that handles it best, then redesign the handoff. The resulting productivity gain is not just faster processing; it can improve the quality of attention given to unusual cases. Yet removing routine work does not guarantee more fulfilling work. Employers can simply raise quotas or expand workloads. The distinction between augmentation and intensification depends on management choices, including whether workers gain discretion, support, and a genuine share in the benefits.
Use AI to widen the search space, not merely speed routine research.
Let machines explore possibilities. AlphaFold2 predicts the structures of nearly 200 million proteins, compared with roughly 170,000 structures solved in the half-century before its arrival. Rather than making researchers irrelevant, the database lets them start with useful predictions and devote more effort to experiments and interpretation.
Insilico Medicine combined AI tools with AlphaFold's protein predictions to identify and design a potential liver-cancer drug candidate in 30 days, a process that can take years. The result was a research candidate, not a proven treatment. Across scientific work, AI can help organize observations, generate hypotheses, and explore candidate designs. At Haverford College, a model trained on records of successful and failed experiments predicted crystal formation more accurately than researchers relying on intuition alone, and its decision tree gave them clues for new hypotheses.
The strongest scientific use is not machine authority but expanded search: algorithms can test patterns across more data and combinations than a research team can inspect manually. That creates room for human interpretation and better-targeted experiments. However, pattern discovery is not the same as causal explanation, and a plausible prediction is not a verified scientific result. The distinction matters especially in medicine, where candidate selection is only one stage in a long chain of validation. AI can make inquiry broader and faster, but evidence still has to earn trust.
Run low-risk trials while funding a few strategic reinventions.
Use two speeds. The book distinguishes no-regrets bets, modest applications that are comparatively easy to deploy, from strategic bets that could reshape a process or a company's position in its industry. A firm might use generative AI to draft marketing material or help staff find information, while separately investing in a larger redesign of product development, underwriting, or inventory planning.
For example, a telecom company used generative AI to help agents answer landlords' questions about tower rental contracts. Agent response time fell by 30 percent, and the user-experience score rose by 66 percent. By contrast, a food company tested AI-supported inventory recommendations that combine market signals with operational constraints. The approach required new workflows, multiple teams, and human approval. The lesson is to learn quickly from practical uses while committing adequate people and capital to the changes that could create lasting differentiation.
This portfolio logic helps avoid two common errors: waiting for certainty while competitors learn, and scattering pilots without a route to business value. Low-risk experiments generate operational knowledge; strategic projects test whether the business can change its underlying process. The difficult part is deciding what deserves scale. A successful demo is not evidence of durable advantage, and an ambitious project can consume resources without fixing a real customer problem. Strong experimentation therefore needs explicit measures of value, risk, adoption, and the conditions under which a trial should stop.
Create human jobs that train, explain, and govern AI systems.
People shape machine behavior. The authors group emerging responsibilities into three roles: trainers improve how AI performs, explainers help people understand its decisions, and sustainers monitor its safety and responsible use. These functions may become dedicated jobs or part of existing roles.
A trainer might improve a customer-service assistant's responses or prepare prompts that produce useful answers. An explainer might investigate why a lending model rejected an application or help choose a more transparent system for a high-stakes decision. A sustainer might test for bias, oversee safety, or define when an AI system needs human intervention. The book also describes machine relations managers, who review AI systems' performance much as human-resources managers evaluate employees. These roles reflect an important point: organizations need people not only to use AI, but also to maintain accountability for how it behaves.
These responsibilities make AI governance part of everyday operations rather than a policy document filed away after launch. They also reveal that technical deployment creates social work: translating decisions, correcting data, and deciding who is accountable. The strongest version of this model gives these workers authority and access to information, not just titles. A caution is that organizations can create impressive-sounding roles without changing incentives or decision rights. Oversight is meaningful only when someone can pause deployment, require remediation, or challenge a system's results.
Treat data, cloud, and skills as operating infrastructure, not side projects.
Build the conditions for reinvention. The book's MELDS framework organizes the work around mindset, experimentation, leadership, digital core, and skills. Its digital core means the connected cloud, data, AI, applications, and security foundations that let a company adapt. Data must be accessible and useful across the organization, not trapped in incompatible silos.
The authors report that 9 percent of more than 1,500 surveyed practitioners worked in organizations they considered leaders in adaptive AI use. Their analysis also describes a 60-40 effect: companies with a strong digital core experienced 60 percent faster revenue growth and 40 percent faster profitability growth. These are the authors' reported findings, not a guarantee for every firm. Their practical advice is to begin with a bounded use of data, then expand access, variety, and speed as the organization develops its capabilities.
The infrastructure argument explains why a model alone rarely transforms a company. Data quality, secure access, employee expertise, and the ability to connect systems determine whether promising technology can reach real work. This resembles complementary investment in economics: a new technology produces more value when organizations also change routines and capabilities around it. The reported growth comparison is suggestive, but it does not by itself prove that the digital core caused the difference. More capable firms may also have stronger leadership, capital, and market positions.
Use augmentation to make work safer, more creative, and more personal.
Three forms of support clarify how AI can extend human capability. Amplification gives workers better analysis or a wider range of options. Interaction lets people use conversational systems to retrieve information or serve customers. Embodiment places AI in physical tools, such as collaborative robots, sensors, and drones.
Autodesk designers used generative software to explore many possible chair designs while keeping practical constraints in view; the selected Elbo Chair required 18 percent less material than the initial model. In manufacturing, collaborative robots can take on heavy or repetitive movements while people handle delicate work. Researchers working with BMW found human-robot teams about 85 percent more productive than either humans or robots alone. These examples show AI changing not only task speed but also design choices, physical demands, and the range of products or services a business can offer.
The value of augmentation becomes clearer when measured in more than output per hour. Reduced injury risk, wider design exploration, and better customer attention are meaningful outcomes in their own right. This broadens productivity from a narrow efficiency metric into a question about capability and work quality. But the same systems can increase monitoring or collect intimate data, as connected retail devices might. A human-centered design must therefore ask both what workers and customers gain and what information, privacy, or control they surrender.
Train people to question, guide, and learn alongside intelligent systems.
Collaboration takes practice. The authors call the abilities needed in human-machine work fusion skills. Their eight examples include redirecting saved time toward distinctly human work, building acceptance responsibly, using judgment when a system is uncertain, asking useful questions, working effectively with assistants, developing an accurate mental model of AI, teaching and learning reciprocally, and continually redesigning processes.
A turbine-maintenance worker with a digital twin illustrates the idea. The system can identify damage, estimate its consequences, and compare repair options. The worker must still ask what the system knows, connect technical findings to cost and operations, and decide what action makes sense. The point is not that every employee must learn to code. Workers need enough understanding to use AI's strengths, notice its limits, and bring their own expertise into decisions.
These skills resemble the practiced fluency people develop with instruments, vehicles, or complex professional tools: competence comes from knowing both what the tool can do and when its output deserves scrutiny. Natural-language interfaces lower the entry barrier, but they do not eliminate the need for domain knowledge. In fact, fluent answers can conceal uncertainty. Training should therefore include critical use, not just prompt technique, and should reward employees for identifying errors rather than treating every correction as resistance.
Give AI boundaries, accountability, and worker agency before scaling it.
Trust requires structure. The book defines responsible AI as intentional design, deployment, and use that creates value while addressing risks such as bias, unreliable outputs, privacy breaches, safety failures, and intellectual-property concerns. Practical safeguards include testing systems continuously, setting limits, explaining decisions where needed, and identifying which decisions require human review.
The authors use the term moral crumple zone for a situation where a person bears blame for a failure produced by a complex automated system. In a ride-hailing example, a driver and passenger faced the consequences of a faulty destination supplied by an app, with no clear way to assign responsibility to the system. Organizations should let workers question recommendations, provide routes for reporting system failures, and make owners accountable. Trust should not mean blind acceptance; it should mean understandable, contestable decisions.
This is a governance problem as much as a technical one. When control is distributed across software, managers, vendors, and frontline workers, responsibility can become hazy precisely when harm occurs. Clear accountability prevents the human operator from becoming a convenient scapegoat for a system they could not inspect or override. At the same time, human review is not automatically fair or accurate. The goal is not to presume that people are infallible, but to make decisions traceable, appealable, and subject to correction.
Analysis
The book's most useful move is to replace the job-versus-machine debate with a question about process design. That reframing directs attention toward task boundaries, human-machine handoffs, training, and decision authority. Its strongest examples show that the result of AI depends less on a model's isolated capability than on the surrounding organization: a complaint system can free staff for unusual cases, while a digital design tool can expose options that human designers might never test. The argument is persuasive as a strategy for creating value, but it should not be mistaken for a forecast that automation will generally preserve jobs or improve job quality. The cases show what some organizations have done, while many figures describe potential impact or company-reported outcomes. Neither establishes how gains are distributed among owners, customers, and workers. Freed time can become more meaningful work, but it can also become higher quotas, reduced staffing, or intensified monitoring. The book's human-centered promise therefore depends on labor choices that technology cannot make on its own.
Its own framework points toward a broader standard for judging AI projects. Beyond speed and cost, leaders should measure whether workers retain agency, whether customers can challenge consequential decisions, and whether errors can be traced to accountable owners. This is especially important as generative systems produce plausible but incorrect answers. Human oversight is valuable only when reviewers have the time, knowledge, and authority to intervene.
The enduring lesson is not simply to combine people and machines. It is to redesign work so that machine scale supports human judgment, while institutions remain responsible for the consequences. AI can widen what an organization is able to attempt. Whether that wider capacity serves people depends on who sets the goals, controls the gains, and bears the risks.
Review Summary
Human + Machine receives mixed reviews, with an average rating of 3.63/5. Some praise its insights on AI's impact on work and business processes, highlighting the concept of human-machine collaboration. Critics argue it lacks depth, originality, and balance, often rehashing known AI concepts. The book is seen as more suitable for executives and those new to AI, offering case studies and a framework for AI adoption. However, some reviewers find it overly optimistic and potentially outdated in the rapidly evolving AI landscape.
People Also Read
Glossary
The missing middle
Human-machine collaboration spaceThe missing middle is the range of work created when people and AI systems collaborate rather than operate as separate replacements. Humans may train, interpret, or oversee systems, while machines extend human analysis, communication, or physical abilities. The concept directs attention away from a simple choice between human work and automation toward designing effective shared work.
Third wave of business transformation
Adaptive process redesignThe authors use this term for a stage of business change in which AI and real-time information enable processes to adapt as conditions change. It follows standardized processes associated with assembly lines and automated processes associated with information technology. The emphasis shifts from speeding up fixed steps to redesigning work around flexible human-machine teams.
MELDS
Organizational AI readiness frameworkMELDS names five areas the authors consider essential to successful AI adoption: mindset, experimentation, leadership, digital core, and skills. It is a guide for changing more than technology. Organizations must rethink work, test ideas rigorously, provide accountable leadership, connect data and systems securely, and prepare people to collaborate effectively with AI.
Fusion skills
Capabilities for shared workFusion skills are abilities that help people and machines work together effectively in changing business processes. The book identifies skills such as asking useful questions of AI, exercising judgment when systems are uncertain, learning reciprocally with intelligent tools, and redesigning work. They do not require every worker to become a machine-learning specialist.
Responsible AI
Trustworthy AI practiceResponsible AI is the intentional design, deployment, and use of AI to create value while addressing risks to people and organizations. The authors emphasize fairness, transparency, explainability, safety, accountability, privacy, security, legal compliance, and sustainability. They argue that these safeguards require ongoing governance and monitoring, not a one-time review before launch.
Digital core
Connected technology foundationA digital core is the connected foundation of cloud services, data, AI, applications, platforms, and security that supports an organization's work. A flexible core makes it easier to access information, connect systems, and adapt processes. In the book's account, fragmented legacy systems and inaccessible data can prevent AI projects from delivering value.
Foundation models
Adaptable pretrained AI modelsFoundation models are machine-learning systems pretrained on large and varied datasets, then adapted to many different tasks. Language models are one prominent type, but models can also work with images, audio, or multiple data types. Organizations can use them directly, customize them with relevant data, or build specialized models for distinctive needs.
Moral crumple zone
Human blamed for system failureA moral crumple zone occurs when a person in a complex automated system absorbs the blame or liability for a failure, even though control was distributed across people and technology. The concept highlights a mismatch between responsibility and actual power. Organizations should make system failures traceable and ensure workers can challenge or correct automated decisions.
Buy, boost, or build
Three model adoption choicesThis decision framework describes three ways an organization can use a foundation model. Buying means using an existing model with limited customization. Boosting means adapting a model with company data or retrieval tools. Building means creating and training a model from the ground up, a costlier route that may offer greater control or differentiation.
FAQ
1. What is Human + Machine: Reimagining Work in the Age of AI by Paul R. Daugherty about?
- Human-machine collaboration focus: The book explores how artificial intelligence, especially generative AI, is transforming work by fostering collaboration between humans and machines, rather than simply replacing human workers.
- Three waves of transformation: It outlines the evolution from mechanization, to automation, and now to AI-driven augmentation, emphasizing the shift toward adaptive, real-time human-machine teams.
- Practical frameworks and guidance: Daugherty introduces actionable frameworks like MELDS to help leaders and organizations adopt AI responsibly and effectively.
- Future of work: The book discusses new roles, skills, and mindsets required for thriving in AI-augmented workplaces, highlighting the importance of ethical considerations and continuous learning.
2. Why should I read Human + Machine by Paul R. Daugherty?
- Dispels AI myths: The book challenges the misconception that AI will simply replace jobs, showing instead how it augments human capabilities and creates new opportunities.
- Actionable strategies: Readers gain practical advice and frameworks, such as MELDS, for implementing AI in organizations and reimagining business processes.
- Ethics and responsibility: Daugherty emphasizes the importance of responsible AI, addressing issues like bias, transparency, and safety.
- Endorsements and credibility: The book is praised by industry leaders and provides up-to-date insights, making it essential for anyone preparing for the future of work.
3. What are the key takeaways from Human + Machine by Paul R. Daugherty?
- The "missing middle": The book introduces the concept of a collaborative space where humans and AI work together, combining their strengths for superior outcomes.
- Generative AI’s impact: It highlights how generative AI expands the possibilities for human-machine collaboration, making AI accessible to more people and functions.
- MELDS framework: Success with AI requires a shift in mindset, structured experimentation, strong leadership, a robust digital core, and the development of new fusion skills.
- Ethical and responsible AI: The book stresses the need for responsible AI practices to ensure fairness, transparency, and trust.
4. What is the "missing middle" concept in Human + Machine and why is it important?
- Definition and significance: The "missing middle" is the collaborative space where humans and AI systems work together, each complementing the other's strengths.
- Business impact: Organizations that fill this middle see significant improvements in productivity, innovation, and customer experience.
- Expanded by generative AI: Generative AI has greatly enlarged this collaborative space, making AI tools accessible to a broader range of employees.
- Real-world examples: Companies like Rio Tinto and Stitch Fix demonstrate the missing middle by combining human expertise with AI-driven recommendations and automation.
5. How does Human + Machine by Paul R. Daugherty describe the evolution of AI and business transformation?
- Three waves explained: The book details the shift from standardized processes (assembly lines), to automated processes (IT-driven automation), and now to adaptive, AI-powered processes.
- Adaptive processes: AI enables flexible, real-time human-machine teams that continuously adapt to new data and changing conditions.
- Generative AI’s role: Generative AI accelerates this third wave by enabling creative, natural language interactions and new forms of collaboration.
- Reimagined business models: The evolution encourages organizations to move beyond automation and toward reimagining entire workflows and business models.
6. What is the MELDS framework in Human + Machine and how should organizations use it?
- Mindset: Encourages breaking from traditional thinking to imagine new ways AI can transform processes, focusing on collaboration.
- Experimentation: Advocates for rigorous, scalable testing and learning from failures to refine AI applications.
- Leadership: Stresses the need for responsible, human-centric leadership that addresses ethical, legal, and psychological challenges.
- Digital Core and Skills: Calls for building a robust digital infrastructure and developing fusion skills that blend human and machine strengths.
7. What are foundation models according to Human + Machine and why are they significant?
- Definition: Foundation models are large, pretrained AI models (like GPT-3, GPT-4, BERT) that can be adapted for a wide range of tasks without task-specific training.
- Capabilities: These models can generate text, images, code, and more, powering applications from chatbots to drug discovery.
- Strategic choices: Organizations must decide whether to buy, customize, or build foundation models, balancing cost, control, and competitive advantage.
- Business impact: Foundation models enable rapid deployment of AI solutions across industries, democratizing access to advanced AI capabilities.
8. What new jobs and roles does Paul R. Daugherty identify in the AI era in Human + Machine?
- Trainers: Individuals who teach AI systems humanlike traits, cultural awareness, and domain knowledge, including personality and localization trainers.
- Explainers: Professionals who interpret AI decisions, ensure transparency, and conduct algorithm forensics to maintain trust.
- Sustainers: Roles focused on ethical compliance, safety, and ongoing monitoring of AI systems to prevent harm.
- Machine relations managers: Oversee AI system performance and integration, ensuring AI tools align with organizational goals.
9. What are the eight "fusion skills" outlined in Human + Machine by Paul R. Daugherty?
- Rehumanizing Time: Redirecting time saved by AI to human-centric tasks like creativity and interpersonal interaction.
- Responsible Normalizing: Shaping public and workplace perceptions of AI to foster acceptance and trust.
- Judgment Integration and Intelligent Interrogation: Applying human judgment to AI outputs and knowing how to ask AI the right questions.
- Bot-Based Empowerment and Holistic Melding: Leveraging AI assistants to extend capabilities and developing mental models for seamless collaboration.
- Reciprocal Apprenticing and Relentless Reimagining: Continuous mutual learning between humans and machines, and the discipline of creating new processes enabled by AI.
10. How does Human + Machine by Paul R. Daugherty suggest organizations should implement and scale AI?
- Lead with value: Start with clear business cases that focus on end-to-end capabilities, not just isolated use cases.
- Develop a secure digital core: Invest in cloud, data integration, and AI infrastructure to support scalable deployment.
- Reskill and reinvent talent: Invest in employee reskilling for new roles and fusion skills, fostering a culture open to change.
- Responsible AI governance: Establish frameworks for ethical AI use, accountability, and risk management, and drive continuous reinvention.
11. What ethical and leadership challenges does Human + Machine by Paul R. Daugherty address regarding AI?
- Responsible AI use: Leaders must ensure AI systems are explainable, unbiased, secure, and compliant with regulations.
- Human agency: Maintaining employee control and decision-making authority is critical when working alongside AI.
- AI governance: Organizations need C-suite accountability and frameworks to manage AI risks and build trust.
- Ethical AI personas: As AI bots become brand ambassadors, their design must align with company values and customer expectations.
12. What is the future outlook and call to action in Human + Machine by Paul R. Daugherty?
- Human-machine symbiosis: The book envisions a future where AI augments human potential, creating new jobs and transforming industries through collaboration.
- Investment in people: Success depends on investing in human skills, culture, and responsible AI governance, not just technology.
- Continuous reinvention: Organizations must embrace relentless reimagining and lifelong learning to keep pace with rapid AI advancements.
- Collective responsibility: Addressing AI risks requires cooperation among businesses, governments, and society to ensure a better future for all.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.