What Is Coming Next

Trying to build for something that does not exist is hard.

That is the sentence I want you to register before any of the predictions in this chapter land, because the sentence is the operating reality I have been living inside for the last three years. CI Web Group has been building infrastructure for a future that has not yet fully arrived — a future where the customer is an AI agent, where the website is a content distribution layer rather than a destination, where the conversion happens inside an LLM conversation, where the unit economics of operations is reorganized around agent-supervision rather than people-management. Every quarter the future arrives a little more. Every quarter the gap between what we have built and what the future will require shifts. We keep building. We keep adjusting. The work does not stop because the destination keeps moving, and that movement is itself the operating condition I am now writing this chapter from inside of.

I want to be honest with you about the epistemic posture of this chapter, because most prediction chapters in CEO memoirs perform a level of certainty that the operator does not actually have. I do not have certainty about what is coming next. I have patterns. I have directions. I have the lived experience of building infrastructure for an emerging environment for three years, which has taught me that the directions matter more than the specific dates and the patterns matter more than the specific products. I am going to give you twelve predictions across six sections. Each prediction is something I am betting on inside CI Web Group right now. Each prediction may be wrong about specific timing or specific implementations. None of them are wrong about the directions they point. The contractor reading this chapter who absorbs the directions will be substantially better positioned for the next three years than the contractor who waits until the specific timing is provable. By the time the timing is provable, the bet has been won by somebody else.

I am betting on these futures even though I cannot prove they are coming, because the cost of being wrong is small and the cost of being late is large. That is the asymmetry that has driven every major operating decision of my career. The asymmetry is what I want you to internalize from this chapter. The directions matter. The exact timing does not.

Voice Becomes the Assistant

The first prediction is the easiest one to defend because it is already happening.

The voice-platform category that defined the late 2010s and early 2020s — Alexa, Google Assistant, Siri — is being absorbed into the AI-assistant category that defines the late 2020s. The standalone voice platforms with their hard-coded skill ecosystems and limited natural-language capability are being replaced by frontier-model AI assistants that operate across voice, text, and increasingly visual surfaces with substantially better understanding and substantially fewer command-and-control constraints. The Echo on the kitchen counter is being replaced by an Alexa-with-Claude-or-GPT-running-underneath-it. The Siri on the iPhone is being replaced by Apple Intelligence. The Google Assistant is being replaced by Gemini. The fragmented voice-platform ecosystem of 2020 is consolidating into a unified AI-assistant ecosystem where the underlying intelligence is a frontier model and the surface where the user encounters it is incidental.

That consolidation matters for trades-industry contractors because it changes what the consumer is actually using when they ask their device a question. The homeowner who said *Alexa, find me an HVAC contractor* in 2020 was operating inside a small, predictable, command-driven interaction model. The homeowner who says the same words in 2027 will be operating inside a frontier-model conversation that can read your website, evaluate your reviews, compare you to three competitors, and recommend a provider — all before the human even touches their phone. The voice surface that used to be a glorified search query is becoming a research-and-decision conversation. The contractor whose digital footprint is not optimized to be evaluated by that conversation is invisible to the homeowner who used to find them through a more passive search.

The convenience is the accelerant. Consumers are not adopting AI assistants because they are technologists. Consumers are adopting AI assistants because the assistants make their lives easier. The friction of typing, scrolling, reading, evaluating, and deciding is replaced by a conversation that does most of that work for them. Convenience is the most powerful force in consumer technology, and convenience is what the AI assistants are delivering. The contractor who is waiting for the AI-adoption curve to slow down is waiting for convenience to stop being valuable to consumers. That is not a thing that is going to happen.

The Customer Becomes the Agent

The second prediction is the deepest one in this chapter.

The customer, increasingly, is not the human. The customer is the agent acting on behalf of the human. That distinction is currently abstract. In three years it will be the operating reality of consumer commerce. The homeowner who needs an HVAC contractor in 2028 will not necessarily be the entity making the calls, evaluating the options, scheduling the appointment, or even authorizing the payment. The homeowner’s AI assistant will. The homeowner will set the parameters — budget range, preferred timing, must-haves and nice-to-haves — and the assistant will execute. The contractor that wins that procurement is not the contractor that won over the human. The contractor that wins is the one whose digital footprint was correctly understood by the AI assistant evaluating the options.

That is the agentic web. It is the term the technology industry has been using for two years to describe the transition we are in the middle of. The transition is from a web of pages that humans navigate to a web of agents that act on behalf of humans. The transition is partial right now. By 2027 it will be substantial. By 2030 it will be the dominant pattern for transactions in many categories of consumer commerce — trades services among them.

AI-to-AI conversions are the operational reality that follows. The homeowner’s AI agent talks to the contractor’s AI agent. The conversation might be measured in milliseconds. The procurement decision is made between two systems. The two humans involved — the homeowner and the contractor — may not be aware the conversation happened at all, or may receive a summary after the fact. That is a different operating environment than the one most contractor businesses have been built around. The contractor whose website is the entity that the homeowner’s AI evaluates is a contractor whose digital footprint is doing the selling on their behalf, twenty-four hours a day, with no salesperson involved. The contractor whose website is not ready to be evaluated by an AI agent is a contractor who is missing the conversation entirely.

AAO — Agent Answer Optimization — stops being an emerging marketing discipline in this future. AAO becomes the entire marketing discipline. The contractor whose digital footprint is optimized for being correctly understood by AI agents is a contractor whose business is structurally aligned with how customers are actually buying. The contractor whose digital footprint is still optimized for human searchers reading blue links is a contractor operating inside an obsolete model. There is no third category. The migration to AAO is not optional. It is the work of the next three years.

Content Cannibalization and Data as Moat

The third prediction is one I think is already happening, even though most agencies and most contractors have not yet absorbed it.

Content cannibalization is the dynamic where AI summaries pull from multiple sources and produce answers that displace the source content’s traffic. The contractor who publishes a blog post explaining the differences between two types of HVAC systems is, increasingly, contributing content that the AI absorbs and repurposes into a summary the homeowner reads without ever clicking through to the contractor’s site. The traffic that used to follow good content does not follow it anymore. The work is being consumed without the consumer arriving at the source. That is cannibalization. It is happening now. It is going to get worse before it stabilizes.

The defense against cannibalization is uniqueness. Specifically, the defense is having entity data that is so unique to your business that an AI cannot simply summarize equivalent content from a competitor. Your specific service area, your specific pricing structure, your specific team credentials, your specific customer outcomes, your specific brand history, your specific operational signals — these are the entity-level data points that AI agents will use to differentiate you from competitors. The contractor whose entity data is generic and indistinguishable from any other contractor in the market is the contractor whose content gets cannibalized into anonymous summaries. The contractor whose entity data is highly unique — specific service guarantees, specific operational metrics, specific local-community ties, specific trust signals that no competitor can credibly claim — is the contractor the AI cites by name.

That is data as moat. The companies whose proprietary data is unique enough to defy simple AI replication become structurally advantaged in the AI era. The companies whose data is generic, fragmented, or unstructured become structurally disadvantaged. Your business, in 2026, is sitting on a data asset whose value depends entirely on whether you have organized it in a way the AI can understand and cite. The contractor with twenty years of customer-relationship history, properly structured, with verifiable outcomes and unique service signatures, is sitting on a moat. The contractor whose data is in scattered spreadsheets and a 2014-era field service management system is sitting on a liability. Both contractors had the same data going in. The difference is the structure.

The work to build the moat is the work to make your data AI-readable, attributable, and verifiable. Structured data on your website that AI agents can ingest. Reviews aggregated and machine-readable. Operational metrics published in formats that AI can cite. Brand and trust signals that are demonstrable rather than claimed. Each of these is a piece of the moat. None of them are optional for the contractor planning to compete in the next three years.

The Collapse of Attribution

The fourth prediction is the one that scares marketing departments the most.

In-LLM conversions are conversions that happen inside an AI conversation, before the consumer ever clicks through to a website or generates the kind of analytics signal that traditional marketing measurement can detect. The homeowner asks their assistant a question. The assistant evaluates the options. The assistant recommends a provider. The homeowner books the appointment. The traditional analytics framework — pageviews, sessions, click-through rates, conversion funnels — sees none of this. The conversion happened inside an LLM. The LLM did not report back to the contractor’s analytics platform. The contractor knows the appointment was booked. The contractor does not know which marketing channel produced it, because the channel itself was a conversation that no analytics platform was inside of.

That dynamic, scaled across the consumer economy, is the collapse of marketing attribution as an industry. Attribution is the discipline of figuring out which marketing channel produced which sale. Attribution depends on being able to track the consumer journey from awareness through conversion. When the journey happens inside an AI conversation that no platform can audit, attribution becomes a guess rather than a measurement. The marketing categories that depend on attribution-based budget allocation — paid search, programmatic advertising, performance marketing — face a structural challenge that goes beyond technique adjustment. The whole budgeting framework of marketing departments is built on attribution. When attribution stops working, budgeting has to rebuild from a different foundation.

The contractor whose marketing budget in 2026 is allocated according to attribution data is allocating dollars based on a measurement system that is increasingly meaningless. That is uncomfortable. It is also true. The contractor who internalizes that fact in 2026 will spend their 2027 marketing budget very differently from the contractor who does not. The categories that hold up are the ones that build durable assets — entity authority, structured content, local-search positioning, AAO-ready architecture. The categories that collapse are the ones that depend on tracking the consumer journey through traditional touchpoints — those touchpoints are increasingly being skipped entirely.

The honest answer about what to measure instead is that the industry has not yet figured it out. We are operating in a transition period where the old measurement system is breaking faster than the new one is being built. CI Web Group is investing in our own internal measurement architecture for AI-mediated conversions. So are some of the more sophisticated agencies in adjacent categories. None of us have a fully solved answer yet. The contractor who is paying attention to which agencies are working on this question is the contractor who will have a better measurement framework in 2027 than the contractor who is still asking their current agency for the same kinds of analytics dashboards their 2018 agency would have produced.

What I can tell you about where tracking and attribution are evolving is the directional shape, even though the specifics are still being built. The new measurement framework is going to be oriented around three things the old framework did not center on. First, AI-citation tracking — the discipline of measuring how often, and in what contexts, your business is being cited inside the AI conversations consumers have when researching your category. The technical capability to measure this is emerging in 2026 and will be substantially more mature by 2028. Second, entity-authority signals — measurable indicators of how strongly the AI agents in your market recognize your business as the canonical answer for specific service categories in specific service areas. Third, persistent-relationship metrics — measurements of how the AI memory in your customer base is accumulating positive or negative associations with your business over time, and what that accumulation predicts about future revenue. None of these three categories existed as measurement disciplines five years ago. All three are emerging right now. The contractor whose agency is investing in all three is the contractor who will have measurement clarity in 2028. The contractor whose agency is still building dashboards on Google Analytics and call-tracking will not.

Tracking and attribution are not collapsing. They are evolving drastically. The contractor who reads the collapse and panics is the contractor who misses the evolution. The framework that comes after the old framework is going to be more useful to the operator who wants to actually understand their customer base, not less, because the new framework measures what is actually happening in the AI-mediated economy rather than measuring proxies for it. The transition is uncomfortable. The destination is better than the place we are leaving.

Managing Agents Instead of People

The fifth prediction has implications for how every business is structured, including yours.

In many roles, by 2028, people will be managing autonomous agents rather than other people. The middle-management category that defined corporate organizational charts for the last fifty years is being reorganized around agent-supervision rather than human-supervision. The skills of being a good manager change when the direct reports are AI systems rather than humans. The org chart changes. The number of humans needed to run a function decreases substantially because each remaining human is now managing a portfolio of agents rather than a portfolio of human direct reports. The leverage is real. The implications are uncomfortable for the millions of middle managers whose jobs are about to be reorganized around a different kind of supervision.

Inside CI Web Group, this transition is already happening. Our AI engineering team supervises agentic workflows. Our vibe-code team works in real-time partnership with AI systems. Our growth division operators are increasingly managing portfolios of automated processes that handle work that used to require human teams. The 38 people we kept through the restructure are not doing what they were doing before. They are doing what comes after. Most of them are now agent-managers in some form. The skills they are developing — evaluating agent output, designing agent prompts, troubleshooting agent failures, deciding when to override an agent’s recommendation — are skills that did not exist as a coherent professional discipline three years ago and that will be the dominant management skill set within five years.

That transition implies the end of certain professional categories in their current form. I want to be careful with this prediction because the journalism about AI-driven job displacement has been wrong about specifics for several years running, and I do not want to hand you a list of professions that will look silly in 2030. What I will say is the structural prediction. Roles that consist primarily of repetitive cognitive work that an AI agent can do faster and more cheaply will mostly be reorganized into much smaller teams supervising AI work. Roles that consist primarily of human judgment, relationship-building, physical work, or creative synthesis will be augmented by AI rather than replaced. The mix in any given profession is what determines which side of the line that profession falls on. Some professions are mostly the first category. Some are mostly the second. Most are some mix. The mix is what is being reorganized.

If you are reading this book and your role consists primarily of repetitive cognitive work, your role is at risk. If your role consists primarily of judgment, relationships, physical work, or creative synthesis, your role is being augmented and you have an opportunity. The honest test is to ask yourself which category most of your hours fall into. Then act on the answer. The lead-the-change posture from Chapter Nine applies. Do not wait for the reorganization to arrive. Lead it inside your own work.

I want to give you a concrete example of what this future looks like in operating practice, because the abstraction above is less useful than the specific is.

At CI Web Group in 2026, our clients are able to communicate directly with a team of trained autonomous agents that have the ability to manage changes to their websites and AI infrastructure in real time, with human oversight from our team backing up the agent layer. The contractor who needs a content update does not wait for an account manager to put the request in a queue and route it to a developer who will execute it next week. The contractor talks to the agent. The agent executes the change. The human supervisor on our side reviews the work, approves it, and intervenes if judgment is required. The cycle that used to take days at most agencies takes minutes at ours. The cycle that used to require multiple human handoffs requires one human review at the end of the work the agents have already done.

That capability exists in production today, with a current cohort of clients, on the HydraOS architecture I described in Chapter Seven. It is a preview of the operating model the entire industry is going to be running on within three to five years. The agency-client relationship in 2028 is not going to look like the agency-client relationship of 2020. The 2020 model was humans on the agency side communicating with humans on the client side, with quarterly business reviews and monthly reporting calls and turnaround times measured in business days. The 2028 model is autonomous agents on the agency side communicating directly with the client, with human supervision running in the background to handle the judgment calls the agents are not yet ready to make on their own, and turnaround times measured in minutes for most work. The contractor whose current agency is operating in the 2020 model is operating with an agency that is not yet ready for the future this chapter is about. The contractor whose agency is operating in something close to the 2028 model already is operating with an agency that has been preparing for the future for several years. The gap between the two is substantial. It will widen every quarter.

The Website Becomes the Avatar

The sixth prediction is about what your business’s digital home actually is, and what it is becoming.

The website-as-static-document model that has defined the web since the 1990s is collapsing. The website was, for most of the web’s history, a destination — a place humans went to read content, click links, and decide what to do next. That model assumed humans were the readers. Humans are no longer the only readers. Increasingly, the AI agents acting on behalf of humans are the primary readers, and the humans visit the website only when they need to interact with it directly rather than when they want to research or evaluate.

The role of a website, in that future, is two things at once. It is a content distribution platform — the canonical source of structured information that AI agents read on behalf of their users. It is also an interactive surface for the rare human visitors who do come to the site directly. Those two functions point in different design directions. The content-distribution function rewards AI-readable structure, comprehensive entity data, machine-verifiable trust signals, and interoperability with the standards AI agents are converging on. The interactive function rewards conversational interfaces, immediate engagement, and the kind of dynamic visual layer that traditional static-page websites do not deliver.

The interactive avatar is the design direction I think the human-facing function moves toward. Rather than navigation menus and hero images and contact forms, the homeowner who visits a contractor website in 2028 may encounter an interactive avatar — a conversational interface that introduces the business, answers questions, schedules service calls, and handles the kind of interactions that today require multiple form fills and email exchanges. The avatar is the website. The website is the avatar. The traditional static-page architecture becomes a backend layer that the avatar reads from and that AI agents continue to read from on the content-distribution side. The visual front-end of the website becomes a conversational interface.

This prediction is the one I am least sure about in terms of timing. The technology to build genuinely useful conversational avatars is arriving in stages, and the consumer expectations for what an avatar should be able to do are still forming. I am confident the direction is right. I am less confident about whether we are two years away or five. What I am confident about is that the contractor whose website is still designed around 2018 user-experience assumptions is the contractor whose digital home is going to feel old to a homeowner who has just talked to a sophisticated AI assistant in their car on the way to the appointment.

The Persistence Question

The seventh prediction is one most contractors are not yet thinking about, and that is going to redefine customer relationships.

Today’s AI assistants have limited persistent memory. The conversation you had with ChatGPT or Claude or Gemini three months ago is mostly not available to the assistant when you start a new conversation. That limitation is being removed. The next generation of AI assistants will have substantially better persistent memory. The assistant will remember the homeowner’s preferences across years, across devices, across service categories. The assistant will remember which contractor served the homeowner well in 2025, which one was a problem in 2024, which one came recommended by a friend in 2026. The assistant will carry forward seven or ten or fifteen years of context about what the homeowner values, who they trust, and who they will not call again.

That changes the customer-relationship dynamics fundamentally. Most contractor business models assume a fresh acquisition cycle every time the homeowner has a new problem. The marketing budget is allocated to win attention from a homeowner who is, in some real sense, starting their evaluation from scratch. In a world with persistent AI memory, that fresh-start assumption breaks. The homeowner is not starting from scratch. The homeowner’s AI is carrying forward the entire history. The contractor who served the homeowner well three years ago has structural advantage that is not visible to the homeowner directly but is operating inside the AI’s recommendation algorithm.

The contractor who served the homeowner badly three years ago has the inverse problem. The bad experience is remembered. The bad experience gets surfaced when the homeowner’s AI is asked to make a new recommendation in the same category. The old assumption that bad experiences fade with time stops applying when the AI is the entity carrying forward the memory. The bad experiences accumulate. The good experiences accumulate. Loyalty becomes structural in a way it has not been in the past, and so does the inverse — customer-acquisition friction caused by past failures becomes durable in a way it has not been.

The contractor planning for this future is the contractor making sure every customer interaction in 2026 is the kind of interaction the AI will remember in 2030. The bad customer experiences are not going to be forgotten. They are going to be encoded in the AI memory of every household the contractor serves and every household connected to those households through shared AI ecosystems. Reputation, in the AI era, becomes a much more durable asset and a much more durable liability than reputation has been in any previous era of consumer commerce.

The Convergence

The final prediction is the largest in scale, and the one I am most certain about even though I am least certain about the specific timing.

Three exponential curves are converging. AI capability is on an exponential curve — frontier model performance is roughly doubling every six to twelve months on most measurable benchmarks, and the progression has not slowed since 2022. Robotics hardware is on an exponential curve — humanoid robotics, autonomous mobile platforms, dexterous manipulation systems are all improving at rates that were not predicted five years ago. Energy density and battery capacity are on an exponential curve — the energy economics that determine which AI deployments are physically possible are improving at a pace that opens new categories of deployment every quarter. Each curve, on its own, is significant. The convergence of the three is the structural event the next decade is going to be defined by.

The convergence produces, by the late 2020s, intelligent physical systems at consumer scale. Robotaxis. Home robots. Autonomous service operations. The Tesla Optimus units, the Figure humanoid platforms, the next generation of dexterous robotics from labs that have not yet shipped consumer products — some combination of these reaches deployment scale within the next three to five years. The trades industry is partly insulated from this because much of trades work is location-specific and judgment-heavy. Some of the trades work is also at risk of being augmented or replaced by intelligent physical systems. Drain cleaning. Routine maintenance inspections. Initial diagnostics. The categories most at risk are the most repetitive and most physically routine. The categories most insulated are the ones requiring judgment, relationship management, and the kind of in-home presence that consumers may continue to prefer to be human-delivered for cultural reasons even after the technology is capable of substituting for it.

The other implication of the convergence is recursive multiplication. AI builds AI. Agents produce agents. Robotic systems assemble other robotic systems. The compounding loop produces capability at a rate that makes prediction beyond two or three years out substantially harder than prediction inside that window. I do not know what 2030 looks like. I have a strong sense of what 2027 looks like. The gap between my confidence at three years and my uncertainty at seven years is the gap that the recursive multiplication produces. Operators who plan against three-year horizons can plan with reasonable confidence. Operators who try to plan against seven-year horizons are guessing. I am operating against the three-year horizon and updating my model continuously as the seven-year horizon clarifies.

The leverage at scale that follows is the framing I want to leave you with. By 2030, I think it is plausible that one operator with the right AI infrastructure and the right operational discipline could outperform a thousand-person traditional services business in the same market. One laptop with AI plus a properly configured agentic stack and a few subject-matter experts could deliver more revenue, with better margins, at higher quality, than the regional operators currently competing in the same category. That is not a metaphor. That is the operating math the convergence produces. The contractor whose competitive analysis is based on traditional headcount-and-revenue ratios is operating from a model that the convergence is about to invalidate. The contractor whose competitive analysis is based on AI-augmented unit economics is operating from a model that the convergence is about to confirm.

Trying to build for that future is hard. It is also the only future I am willing to spend the next decade of my career building for, because the alternative — building for the world we are leaving — is building for a market that is shrinking every quarter.

The Policy Environment

There is one more piece of the picture I owe you before this chapter closes.

Everything I have predicted in this chapter assumes a policy environment where AI deployment continues developing on the schedule the labs and platforms are setting. That assumption is partial. AI deployment is also being shaped, increasingly, by the regulatory and policy environment that governs how the technology can be built, sold, and used. That environment is not yet settled. The European Union’s AI Act is in force and being interpreted. State-level AI regulation in California, Texas, New York, and elsewhere is being written and rewritten as I am writing this book. Federal policy in the United States is being negotiated across multiple agencies and committees. Export controls on frontier-model technology are shaping which capabilities are available in which markets. The contractor reading this book in 2026 is operating in a regulatory environment that will look meaningfully different by 2028, and the difference is going to affect every business that depends on AI infrastructure.

I want to be honest about my epistemic position on policy. I do not have specific regulatory predictions. The reason I do not have them is that the policy environment is being shaped right now, in real time, by political and institutional dynamics that operators in my position do not have visibility into. The contractor advisor who is confidently predicting what AI policy will look like in 2028 is making predictions outside their actual expertise. So am I, if I attempt the same. What I can tell you is the directional argument. Policy and regulation are going to shape AI deployment in ways that affect every business reading this book. The specific shape of the regulation is being negotiated as I write this. The contractor who is paying attention to the regulatory environment in their state and category will navigate the next three years better than the contractor who is not. That is the durable framing. The specifics will be your local attorney’s job to interpret as the regulations actually land.

I will say one thing about the policy environment that I do feel confident about, because it is observable in the patterns of regulatory action over the last two years. The regulatory environment is going to favor businesses that have built AI infrastructure transparently, with clear human oversight, with auditable decision-making, and with documented safety practices. The regulatory environment is going to disfavor businesses that have deployed AI opaquely, without oversight, without auditability, and without documented practices. CI Web Group’s HydraOS architecture, with the autonomous-agent-team and human-oversight model I described earlier, is structured for the regulatory environment that is arriving. Most agency stacks are not. The contractor whose agency partner has built for transparency and human oversight is the contractor whose business is going to navigate the policy environment without disruption. The contractor whose agency has not is going to face costly retrofitting work as regulations land. The cost asymmetry is the same as the AAO cost asymmetry. Acting now is cheaper than acting later. That is the only thing I will predict about policy with confidence.

Twelve predictions across nine sections.

Voice platforms consolidate into AI assistants. Convenience drives consumer adoption past every previous technology curve. The customer becomes the agent. AAO becomes the dominant marketing discipline. AI-to-AI conversions become the procurement reality. Content gets cannibalized unless entity data is unique enough to defy summary. Data becomes the moat for businesses with structured proprietary information. In-LLM conversions break attribution-based marketing measurement, and tracking and attribution evolve drastically into a new framework oriented around AI-citation tracking, entity-authority signals, and persistent-relationship metrics. People manage autonomous agents instead of other people, and clients communicate directly with autonomous-agent teams with human oversight rather than through traditional agency account-management chains. Some professional categories are reorganized into smaller agent-supervision teams. Websites become content distribution platforms with avatar interfaces. Persistent AI memory makes reputation a substantially more durable asset and liability than ever before. Three exponentials converge to produce intelligent physical systems at consumer scale. Recursive multiplication makes seven-year prediction unreliable and three-year prediction the operating horizon. One operator with AI leverage outperforms a thousand-person traditional business by 2030. Policy and regulation will shape all of the above in ways that favor transparency and human oversight, and the businesses that have built for that policy environment in advance will navigate the regulations without the costly retrofitting that will face the businesses that have not.

Some of these predictions will be wrong. None of them are wrong about the directions they point. The contractor reading this chapter who absorbs the directions has a window to act before the predictions become consensus. By the time the predictions are consensus, the bet has been won by somebody who acted on them earlier.

There is one chapter left in this book.

Some of these predictions will be wrong. None of them are wrong about the directions they point.

It is the chapter I have been quietly writing toward since the Foreword. It is the chapter where the operational predictions in this book meet the personal stakes that gave the book its name and its dedication. It is about my father. It is about the technology that could not save him in 2019 and the technology that might save the next person like him in 2030. It is about why AI is scary and why AI matters anyway. It is about what it means to lean in responsibly to a technology that can hurt as much as help.

It is the chapter I owe you, and the chapter I owe him.

Read on.