Five Predictions for 2026

You can no longer say with a straight face that your organization knows how to make things, build things, and grow. You know how to buy things. You know how to present things. You know how to talk about transformation. But if you had to start from a blank site and build a plant, a supply chain, and an integrated business that could actually move at the speed of events, could you do it without recreating the same staircase of delay that is slowly killing you now. 2025 was the year of agent washing. Everyone discovered the word “agent” and glued it onto dashboards, chatbots, and tired workflows. Very few changed the architecture of who is allowed to act and how fast. 2026 is different. This is the year architecture calls your bluff. The year capital quietly walks away from first generation AI and SaaS as overhead, and toward second generation, agentic architectures that can prove they remove burden and latency. The year you are forced to admit that efficiency is not another round of cost cuts. It is your capacity to navigate complexity at the speed required to make, build, and grow. Here are five predictions about how that will play out. If you sit in the CEO, COO, or CFO chair, these are not trends to watch. They are decisions you either make deliberately or have made for you. Prediction 1. Capital Walks Away from First Generation AI and SaaS as a Tax For the last five years, almost any project with “AI” on the cover slide could find money. First generation AI sold you comfort. It told you there was meaning in one more predictive score, one more dashboard, one more “insight” about last quarter. In 2026, that mood breaks. Quietly at first. Ruthlessly in the end. Boards and CFOs will start asking the only three questions that matter about every AI and SaaS dollar. 1. How many minutes of decision latency does this remove. 2. How many human handoffs does this eliminate. 3. How much cognitive burden does this take off the frontline. If the project cannot answer in real numbers, not stories, the renewal gets cut. The roadmap gets moved “beyond the planning horizon”. The vendor gets one last polite email. The pattern looks like this. Budgets move away from first generation AI that behaves like a rearview mirror. It consolidates and decorates the past, right at the moment your operations need to lean into the future. Budget

also moves away from SaaS sprawl that adds icons to the desktop while quietly stealing time, attention, and energy from the people closest to the work. Money moves toward second generation, agentic architectures. Systems that: • Hold permission, context, and memory. • Live on the edge with your people, not up in the tower with your dashboards. • Can show a simple, causal chain from their existence to cycle time, quality, safety, and cash. You will know you are on the wrong side of this shift if your AI portfolio produces more slideware than step removal. You will know you are on the right side when your operators, schedulers, planners, and sales teams can point to one agent and say, without jargon, “that thing took this work off my back, and nobody asked me to learn another system to get it.” This is not an ideology shift. It is a capital allocation shift. First generation AI comforts you that you own the latest tools. Second generation AI changes the architecture of work. In 2026, the money starts to know the difference. Prediction 2. The Architecture of Burden Is Exposed. Menus Lose and Taskless Surfaces Begin to Win For thirty years we treated menu driven interfaces as progress. ERP, CRM, MES, HCM, PLM, you know the alphabet. The story was always the same. Standardize the process. Put it in a system. Train people to follow the screens. Measure compliance. You did all of that. Now look around. Your most experienced people spend their days acting as human middleware between systems that do not talk to each other. New hires take six months to learn the map of menus and passwords before they are useful. Managers prepare reports about reports. Everyone feels busy. Few feel effective. In 2026, this quietly gets named for what it is. An architecture of burden. Leaders begin to measure burden instead of celebrating it in glossy adoption reports. They look at: • Clicks per transaction. • Systems touched per role per shift. • Minutes of every day spent navigating, reconciling, and correcting systems rather than doing the work the company actually exists to do. The uncomfortable conclusion emerges.

We did not just digitize the work. We digitized the friction. We industrialized distraction and called it transformation. The winners start to design the opposite. Not more apps. Not prettier menus. One simple surface, many systems. A taskless, agentic layer sits on top of your legacy stack. It uses the plumbing of ERP, MES, CMMS, LIMS, CRM, but it does not ask humans to live down there. An operator, planner, or salesperson interacts with one guided surface that: • Already knows who they are. • Already understands the context of this asset, order, student, or customer. • Preassembles the transaction, checks constraints, and proposes the right next action. The human confirms, corrects, or escalates. They do not hunt through seventeen screens to find a field someone forgot to map. I expect at least one major company to make this explicit in 2026. They will admit, in plain language, that they wrote off a massive ERP or SaaS front end and kept the records of truth while replacing the interface with an agentic layer that removed thousands of hours of navigation from the system. Analysts will catch up. A new distinction appears in coverage. Software that adds burden versus software that absorbs it. The former starts trading at a discount. The latter gets treated as real leverage. If you are still buying menus instead of agents by 2026, you are not modernizing. You are adding layers of rust to a machine that can already barely turn. Prediction 3. Decision Latency and the Permission Staircase Become Board Discussions Every CEO talks about speed. Very few can draw how a decision actually moves through their organization. In 2026, that gap starts to close. The companies that matter will know their permission staircase the way they once knew their org chart, and they will talk about it relentlessly in the boardroom. Here is the staircase. • Step 0. A signal appears at the edge. A sensor reading goes out of range, a quality check fails, a customer threatens to churn, a student’s mastery drifts, a supplier misses a window. • Steps 1 to N. Each approval, review, or escalation required before someone can bind the company to an action. • Top step. The minimum level where authority and risk appetite are allowed to intersect.

Right now, most of that staircase is invisible. You feel it as delay. You experience it as “we are working the issue.” Your people experience it as frustration and learned helplessness. In 2026, leading COOs and CFOs will put three artifacts in front of their boards, and they will talk through them in plain language. 1. A decision latency curve for critical decision classes. Safety corrections. Quality escapes. Maintenance interventions. Pricing moves. Credit approvals. Customer remediation. How long from signal to committed action, both median and worst case. 2. A permission staircase map. How many steps each decision class climbs. Which steps are genuinely required for control and which exist because no one ever took a red pen to them. 3. An agent participation map. Where agentic systems already own or share steps inside pre agreed guardrails, and where human escalation is still required for legal, ethical, or reputational reasons. This is where second generation AI moves out of the lab and into the operating model. You will assign agents defined steps on the staircase where they can act without human permission, because you already negotiated the boundaries in daylight. For example: • An agent can automatically schedule and execute certain maintenance tasks within a spend limit when specific risk patterns appear. • An agent can issue concessions or credits up to a threshold when defined customer conditions are met. • An agent can adjust production schedules locally within capacity limits to absorb shocks without waiting on corporate S&OP. The human still matters. They define the rules of engagement, set the guardrails, and own the questions where the organization must look itself in the mirror. The prediction is simple. By the end of 2026, the companies that survive volatility best will be the ones that can answer three questions in under five minutes, in a board discussion, without hiding behind jargon. • Show me, on one page, how a real safety or quality decision moves from sensor to action in this company. • Show me exactly where we have chosen to keep humans in the loop and why. • Show me one staircase where agents have already cut latency without increasing risk. If you cannot do that, you are not running an AI strategy. You are running a latency strategy and pretending it is about technology. Prediction 4. Agents That Shape Outcomes Displace Fake Agents and Causal AI’s Contribution to Agency Becomes Better Understood

2025 flooded the world with fake agents. Icons and chat windows that promised to “assist” but did little more than wrap old workflows in new language. Bots that filled forms, moved tickets, and answered FAQs while carefully avoiding the one thing that defines a real agent. Direct, accountable influence on outcomes. In 2026, that gap closes. Agents that shape outcomes displace fake agents that decorate process. And the reason is simple. Boards and operators finally understand that there is no real agency without causality. Your plants, networks, and customer journeys already generate more data than any human team can keep up with. First generation AI used that data to describe what happened. Sometimes it even predicted what might happen next if the world conveniently behaved like the past. It lived on the bottom rungs of the causal ladder. Seeing patterns, not levers. Causal AI climbs higher. It can answer three harder questions. • What is actually driving this outcome, not just moving alongside it. • What will happen if we intervene here instead of there, now instead of later. • Which actions we should never take, no matter what a correlation chart suggests. Real agents in 2026 will be built on this kind of causal understanding. They will sit inside your operating model with clear mandates, clear guardrails, and clear accountability for moving specific outcomes, not just tickets. An automated scientist is one of those agents. It is what causal; agentic AI looks like when it shows up for work. It does five things, relentlessly. 1. Watches for drift in the variables that matter. Yield, scrap, energy, mastery, cycle time, unplanned downtime, abandonment, slippage. 2. Proposes causal hypotheses instead of correlation gossip. It uses a real model of how your system behaves, not just “things that moved together in a spreadsheet.” 3. Designs low risk experiments with predicted effect sizes and clear guardrails. Which input to tweak, which cohort to touch, which schedule to adjust, what downside to cap. 4. Runs those tests within pre-approved authority. It does not seek permission every time. You gave it permission once, on the staircase, under defined rules. 5. Updates guidance, parameters, or playbooks when the evidence is strong enough, and tells you what it changed and why. Continuous improvement stops being a slow, sociable ritual. It becomes a live loop driven by causal test and learn. Engineers, teachers, planners, and supervisors gain something they have not had in years. They get out of the report making business and back into the cause-and-effect business. They spend their time asking better questions. What outcome matters most now. What risks are we truly willing to accept. What tradeoffs will we no longer tolerate.

The automated scientist grinds. Humans decide. Causal AI supplies the levers, the confidence intervals, and the boundaries where agents should never act. That is what it means for an agent to shape outcomes instead of impersonating a clever helpdesk. Real agents stand on a causal model of your system and take bounded actions that move it. Fake agents stand on a workflow and move messages around. In a world of reshoring and new capital projects, this difference will matter. The old model opens a new facility with thick binders of “best practices” and freezes them in place. The new model opens with a causal model and an automated scientist already embedded, learning the behavior of that specific plant, workforce, and supply base from day one, and adjusting in real time inside the guardrails you set. By the end of 2026, serious boards will ask for one simple proof. Show us one loop in our business where AI does more than report and more than assist. Show us where a causal model plus an agent is actively discovering how to run this system better, under our rules, every day, and where you are willing to let it shape the outcome. If your answer is a slide with “labs” and “pilots” and a row of fake agents that move tickets but never touch the P&L, you are volunteering to fund someone else’s learning curve instead of your own. Prediction 5. The Efficiency Reckoning. We Realize We Forgot How to Make, Build, and Grow For years, “efficiency” has been code for “do the same with fewer people” or “hit the quarterly target no matter what it does to next year.” In 2026, a harsher definition takes hold. Efficiency becomes your ability to navigate complexity and decision latency at the speed required to make, build, and grow in a world that is reshoring, rearming, and reconfiguring at the same time. Three forces collide. • Reshoring and industrial policy are pulling production, energy, and critical supply chains closer to home. • Capital abundance is flowing into “transformation”, but very little of that money actually touches the operating muscles that move steel, electrons, and goods. • Mergers and acquisitions have been used as a substitute for organic growth, without building a repeatable muscle for integrating architectures and operating models. The result is a strange kind of weakness.

On paper, you own more assets than ever. You have more plants, more SKUs, more systems, more “platforms”, more data. In practice, you feel slower, more constrained, more fragile. The reason is simple. We spent thirty years optimizing around making. Optimizing the spreadsheets, the contracts, the presentations about operations, the compliance, the stack of tools sitting between the real work and the real customer. In the process, many large companies forgot how to make things with conviction and how to grow without buying someone else’s story. Your 20 20 60 model is the cleanest way to see this. • Twenty percent of productivity is doing the right things. Choosing the right products, markets, sites, and acquisitions. • Twenty percent is doing those things right. Process discipline, engineering, quality, and integration. The craft. • Sixty percent is staying focused on those things without being dragged sideways by structural distraction. Most executives talk endlessly about the first twenty. Strategy offsites, M&A decks, capital plans. Some still care about the second twenty. Craft and discipline. Almost nobody owns the sixty. That is where the architecture of burden and the permission staircase live. That is where Gen1 AI and SaaS quietly erode your ability to operate by turning every role into a part time systems integrator. That is where complexity and latency pile up until your people cannot tell whether they are working for the customer or for the internal machine. In 2026, the gap between companies that repair the sixty and those that ignore it will explode into the open. You will see it in reshoring. The lazy version of reshoring builds a new plant and drops the same stack of approvals, menus, and reports on top of it. The productive version designs the staircase and the agentic layer first, then wraps concrete and steel around it. You will see it in M&A. The lazy version buys a company for its growth and then suffocates it inside the parent’s stack. The productive version treats architecture and decision speed as assets. If the acquired business has a better staircase and a cleaner edge operating model, the parent integrates into that pattern rather than forcing the smaller company into its friction.

You will see it in talent. The best operators, engineers, architects, and product leaders will quietly migrate to the places where it is possible to do excellent work without being buried under nonsense. The others will keep posting “AI transformation” announcements while their best people quietly leave. By the end of 2026, efficiency will have a new, unforgiving measurement. • How many decisions that matter can move from sensing to acting in hours, not weeks. • How much of your workforce’s cognitive capacity is pointed at the customer, the product, and the asset, rather than at the tools that sit in between. • How quickly you can translate industrial policy and capital incentives into real, functioning, reliable capacity without three years of death by meeting. If your executives can speak fluently about AI, M&A, and “digital” but cannot explain how a real decision moves on a Tuesday night in a real plant, you have your answer. You have not solved efficiency. You have renamed the problem. What To Do Before 2026 Gets Away from You Predictions are entertainment unless they turn into an operating agenda. Here is where you start, without another committee. 1. Draw one permission staircase o Pick a decision that matters. A safety intervention. A quality escape. A major customer concession. A pricing move. o Map every step from signal to action. Name the handoffs, the delays, and the “we always do it this way” points. o Decide where agents could safely take a step under clear rules, and where human judgment must remain. 2. Find the worst piece of your architecture of burden o Identify the single role in your company that touches the most systems in a day. o Stand behind them for an hour and count the clicks. o Commit, in writing, to remove half of that burden within twelve months using an agentic surface that sits above the stack you already own. 3. Deploy one automated scientist in anger o Choose one variable that really matters. Yield on a critical line. Mastery in a grade. Uptime on a bottleneck asset. o Give an automated scientist enough permission and data to watch, propose tests, and tune under guardrails. o Judge it by outcomes, not presentations. 4. Redefine efficiency with your board o Stop reporting “number of AI projects” and “systems implemented” as progress. o Start reporting decision latency, burden removal, and focus protection as core productivity metrics.

5. Force every AI and SaaS dollar through the simple test o If this spend does not materially shrink the permission staircase, absorb burden, or empower an automated scientist to discover better ways of operating, it is likely firstgeneration comfort, not second-generation leverage. Fund accordingly. You do not have to believe these predictions. Your competitors only have to act on them. The companies that will look back on 2026 with quiet satisfaction will not be the ones that shouted the loudest about AI. They will be the ones that rebuilt the architecture of how work, permission, and learning move through their organizations, and who used agentic AI to strip away the rust that has accumulated over thirty years of well-intentioned complexity. Everyone else will still be admiring their dashboards. References This argument draws on and adapts our prior work rather than citing it verbatim, including Enough Intelligence. Shaping Destiny Without Digital Gods, The Line Between First-Generation AI and Second-Generation AI, The Architecture of Permission, The Question Engine, Data as the Comfort Engine of Second Place, the emerging Voter’s Guide to AI material, and the internal One Degree for Everyone and Everything and Architecture of Conversion white papers.