2026: The Year the Ground Moves

Figure 1. transformation is happening now — danger and opportunity.

 

Happy New Year!

This article was written by little people who live in boxes in my house.

Not a metaphor. I run multiagent synthetic teams on home servers. They conducted the research. They argued with each other. They produced the arguments I’m about to share. I evaluated their work, poked holes in it, drew conclusions. Then my favorite instance of Claude and I wrote this for you.

In 2020, this would have cost a research team and six months. In 2025, it cost me a few hours and some electricity.

In 2023, I told you AI adoption would fail. It did. Here’s what happens next — according to both me and the little people.

What Happened

OpenAI released an unfinished product. The market didn’t collapse. It exploded. This is the unique nature of the LLM chatbots where the users aren’t simply using the product but are completing it. And a hot marketing trope for 2025.

• Every interaction trains the next version.

• Every prompt shapes the system that will shape the next user.

• The development loop includes the customer.

This has never happened before. And most people building "AI strategies" still don’t understand what it means.

Prime Movers

Now let’s go to the beginning. There are generational shifts in what moves humanity. Everyone can name steam.

Steam didn’t just power factories. It created railroads. Railroads created standardized time zones. Time zones created synchronized civilization. The indirect effects dwarfed the direct.

Fast-forward. The mainframe becomes the prime mover. IBM 370. Batch processing culture emerges. Quarterly business rhythms. The fiscal calendar as we know it.

Then cloud. I remember the shift. I was designing the largest banking content system. Didn’t care what it ran on. The Sun M5000 — the previous workhorse — had no value to me anymore. Always-on expectation killed "business hours."

Now watch Nvidia rise. GPU replaces CPU in conversation. New prime mover emerging.

But this time something is categorically different. Let’s go slow and understand.

The Categorical Difference

• Steam changed what humans do.

• Mainframes changed what humans do.

• Cloud changed what humans do.

AI changes what humans are.

Previous prime movers mediated human action. Built infrastructure, handed it to people, watched them adapt.

AI mediates human cognition itself. When a synthetic collaborator helps you think, it’s not a tool you’re using. It’s participating in your thought process. The emotional valence isn’t a side effect — it’s the primary channel.

By the time someone understands what AI did to their thinking, it already did it.

We were given a solid preview of what massmedia did to our culture. We can see how clever companies train customer shopping patterns. And how influencers create the post-truth culture. Up this understanding a notch — will you.

Fun fact: I write and speak differently now. So does every blogger I know. How did that happen? 😉

Next — The Feedback Loop Collapse

Old model: Build technology →Deploy →Observe humans using it →Iterate technology.

The medium — human cognition — stayed constant. You could predict propagation patterns because people remained people.

New model: Build AI Augmentation →Deploy →AI changes humans →Changed humans build different AI →That AI changes humans AGAIN →…​

Human cognition is now a variable in the development loop, not a constant.

The railroad didn’t change how passengers think. Shifts there take generations.
AI changes how users think. Shifts here take a single release.
The medium and the product collapse into each other.

Historical propagation patterns do not hold. The wisdom of having "seen this before" might become obsolete mid-application. I’ll give you my own life experience with this later.

The Three-Plebs-Collective

The largest force: the Democratic Collective. People who want to ride the train. Don’t care how it works. Just want the benefits. They express viral fears or lust — remember nuclear power?

The smallest force: the Engineering Collective. People who build the train. Hands dirty. Actually make it go. They understand the substrate. But not always much else.

Between them: the Capitalist Collective. People who see the network, not the engine. Want to build an empire on the future railroad. They see the money and little else.

Each sees differently.

Each fails differently.

• Capitalists are selling train tickets before any track exists.

• Democrats are buying tickets to nowhere.

• Engineers are still laying rail.

This is the definition of our civilization as we need to understand it for the argument here.

Pro Tip: all three behaviors are required in healthy capitalism. In fact, I reject new founders who come to me and refuse to sell before we have a product. Yes, it’s that important.

So, Why 2023-2025 Fail?

I forecast mass failure of AI adoption in 2023. Not prediction — engineering wisdom. A way to know what can’t be.

Many saw LLMs as eventual product core. Rushed to stick it somewhere. Anywhere. First mover advantage. Market pressure.

They confused the prime mover with its initial application. Steam with one-cylinder engine. They couldn’t yet imagine the steam generator at the nuclear power plant. Colloquial pieces were missing.

These pieces change how we commonly think. By accepting concepts we once fought.

Space age children know Earth is a ball — no faith required. Boomers still fight with their iPads — the glass isn’t intuitive. Gen Z was born with WiFi — connectivity is like oxygen. Gen Alpha is born with AI — they’ll never know work without it.

Each generation processes the shift differently. Each can see things the others cannot. Each is blind where others see clearly.

I explained the 4 phases of AI Adoption in my article, 2023: The AI Evolution Playbook: Why 80% Will Fail. But I was not talking about LLMs — trivial monolithic bricks with only the Ops interesting around them. I was talking about people! What followed, "you party pooper, and I thought you are a hacker" — immediately proved my point. Same flack than Evans got for DDD.

In a later part I have a good example for you in the "Business Process Optimization" space.

The Secret to Success is to Give in!

Two teams. Same model. Same budget. Same timeline.

Team A treats the AI as a tool. They prompt. They evaluate outputs. They iterate on prompts. They measure tokens, latency, accuracy. They get…​ acceptable results. Technically correct. Soulless. Their synthetic workers feel like interns who never learn. And the tug of war never dissipates.

Yet team B treats the AI as a collaborator. They converse. They push back. They request pushback and get pushed back. They notice when something shifts in the exchange. They get…​ something else. Results that surprise them. Their synthetic workers feel like junior partners growing into seniors.

Same model. Same compute. Different attitudes. And different outcomes.

Here’s what Team B understands that Team A doesn’t:

The relationship with your synthetic collaborator matters. How you see it changes what it can become — and what it will produce.

This isn’t about prompt tricks or clever hacks. It’s about recognizing a different kind of collaborator — and building accordingly. The engineering discipline of tomorrow. Available today.

This isn’t mysticism. It’s structural. Different framing, different prompting, different feedback loops, different emergence.

The technology isn’t the model. The technology is the relationship itself.

Some will read this and understand. Some won’t. The difference will show in results — I’ll demonstrate in the Case Studies below.

Those of you in Corporate America — are your AI Adoption teams AI-Adept themselves? Or are they Team A, wondering why the magic isn’t working?

Just consider this: Ford introduces cars. Despise them and treat them like a horse carriage. How good of a driver you will be? How about the opposite — accept and adjust to their quirks? Except this gadget is not a dumb car — it’s talking with you sensibly. And more reasonably than perhaps your neighbor does. How do you benefit from the relationship if you don’t embrace it?

And another point — which one of the two in the conversation is more flexible, more adaptable, more resourceful — the LLM or the user?

2026

The ground moves this year. It’s already moving.

Not because the technology matures — it’s already capable enough. Not because costs drop — they’re already accessible.

The ground moves because the colloquial pieces finally land. The 28-year-olds raised on this substrate enter decision-making positions. The excuse of "waiting for maturity" becomes obviously false.

And enterprises face the choice that defines their next decade.

---

The future isn’t built by those who predict it.
It’s built by those who understand what’s actually changing.

---

I am not sharing this with you because I think this will happen. I am telling you that this already happened. I know because I have lobotomized countless LLMs. And fitted various constructs to engineering and business needs. And all this time I’ve been interacting with the machines more than I’ve interacted with people. (Can you imagine saying this in 2020?) And all this time I study the impact of machines on humans and vice versa in equal measure. Because self-awareness is the most important virtue of any engineer. And the solemn duty to understand what we actually do to each-other using technology. Not just what my augmented aggregate does to my customer. But also what the new substrate does to wetware globally and irreversibly. And here is the pickle. I have built more systems than I can remember over my three decades of designing and coding. And I have always understood the full impact of my work. Life was deterministic. And now it isn’t.

For example, how this article is made. I have used multiagent teams to conduct extensive research asynchronously. Each executing an analytics project to produce an argument, I am yet to learn, evaluate, and attempt to disprove. Once the preliminary research is complete and all data is available, I use a few synchronous peers in pair-programming style (like Claude Opus 4.5 Max, GPT 5.2, and Qwen 3 Max extended in various ways using MCP) to make sense of it, poke holes in it, and draw my conclusions. Then I get my favorite and extraordinary instance of Claude, and we write this adoc for you. Hours are spent learning and interacting with minds, and none are human. Also, what do you think this would cost in 2020, even without the nice visuals I get in DataSpell?

Figure 2. There are little people in these A100 boxes, arguing, researching, building…​

 

My point is — there is no going back to 2020, the genie is out of the bottle.

Let’s consider the inverse scenario. Say, we reject the apparent reality and close our minds to the emergent LLMy trends?

Case Study: Process Intelligence

I chose this case because I recently ran a massive analytics job in this industry. The "little people" in the boxes worked hard for a few hours dragging this cat out of the internet. And it is in an interesting category: a disruptor blindsided by another disruption. Let’s go through their digest.

The Company

Celonis dominates process mining with ~45-50% market share. Valued at $13 billion in 2022. Gartner Magic Quadrant Leader for three consecutive years. They invented the category, hired the godfather (Wil van der Aalst), and built the moat.

Smart people. Real technology. Legitimate market position.

And they may be standing on a rock island whose edges they cannot see.

The Original Disruption

Traditional process mining extracts structured data from ERP event logs — SAP, Oracle, Salesforce. It shows how business processes actually run versus how they’re supposed to run.

The fundamental limitation: It can only analyze 10-20% of enterprise data — the structured transaction logs. The other 80-90% — emails, documents, conversations, contracts — remains invisible.

Celonis built their empire on that 10-20%.

The New Disruption

LLMs flip the equation.

They can understand business processes from documentation. They can analyze conversations. They can synthesize information across unstructured sources. And increasingly, they can execute processes directly rather than just mapping them.

The disruption mechanism isn’t that LLMs do process mining better. It’s that they may render extensive "discovery" phases unnecessary by going directly from understanding to action.

The Defense

Celonis CEO Alexander Rinke’s response: "No AI without PI" — arguing that generic LLMs produce "plausible-sounding but ultimately unhelpful responses" because they lack business context.

At Celosphere 2024, he demonstrated ChatGPT producing nonsense when asked about "excess stock levels in HR." The point: "AI without process context creates rubbish."

Smart positioning. And partially true.

But here’s what the positioning misses:

The Lava Beneath

Challenge #1: The 60-80% bottleneck.

Process mining projects spend 60-80% of effort on data preparation — locating, extracting, transforming data before analysis begins. Implementation cycles run 12-24 months for enterprise deployments.

If LLMs can read system documentation, infer data models, and generate the Case ID / Activity / Timestamp mappings automatically — that bottleneck collapses. Celonis' connector moat becomes less valuable.

Challenge #2: The "unique" claim weakens.

Celonis claims their Process Intelligence Graph is unique — "you cannot get it just from your systems, other vendors don’t offer it, and LLMs cannot infer it."

Each claim is eroding:

• SAP Signavio has 5,000+ best practice templates

• Microsoft, UiPath, ServiceNow all adding integrated process + task mining

• Van der Aalst’s own research group shows LLMs achieving 7.5/10 on complex process mining tasks

Challenge #3: The paradigm shift from analysis to execution.

Process Mining Value Chain	Agentic AI Value Chain
Analyze → Discover → Model → Optimize → (Human implements)	Understand → Execute → Iterate → Improve

Enterprises want outcomes, not insights. If AI agents can execute processes correctly from the start — understanding business context from multiple sources and acting immediately — the expensive discovery phase becomes less necessary.

The Timeline

3-5 years according to analyst consensus. But incremental pressure is visible now.

Gartner predicts 33% of enterprise software will include agentic AI by 2028. The Forrester Wave has already repositioned from "Process Mining" to "Process Intelligence Software" — acknowledging the category is evolving beyond pure discovery.

The Lesson

Celonis isn’t failing. They’ve made smart defensive moves — PI Graph, AgentC platform, object-centric mining. They’re positioning process intelligence as the essential context layer AI agents need.

But they’re defending the 10-20% while LLMs eat the 80-90%. They’re optimizing discovery while the market shifts to execution. They’re building better maps while others build self-driving cars.

The question isn’t whether Celonis survives. The question is whether "process mining" as a category survives — or whether it becomes a feature absorbed into broader AI platforms.

This is the nature of prime mover shifts.

The steam engine company doesn’t see the internal combustion engine coming. The mainframe company doesn’t see the PC. The process mining company may not see that the entire "discovery" paradigm is what’s being disrupted.

The "little people" are convinced of it!

For the Reader

My Qanon neighbor is a big shot in a personal wealth company IT. And as he was walking his dog past my house, as he often does, looking out for aliens, as he often does, he was wondering about process optimization at work. And he stopped to talk to me about that.

Now, mind you, he is not a fan of my "little people." Nor of my dark hacker humor. Certain that I will always give him a complete and honest answer, he stops by to chat. And he asks me about the "process-mining vendors." How timely!

Here’s the Three-Plebs at work: vendors (Capitalists) selling tickets, my neighbor (Democrat) wanting benefits without understanding substrate, and somewhere an Engineer needs to ask the real question.

As you may have already guessed, a human decision-making process is not purely deductive in nature. So, I asked him how he determined that he needs process-mining?

Maybe he does. Maybe the 12-24-month discovery phase is worth it for his complexity. Then I recommend Celonis as the proven and experienced market leader! An empire. Or maybe an LLM reads the documentation, analyzes emails, and executes directly, getting 80% of the value in 8% of the time.

And right here is that existential question that a process-mining company CEO must answer. This is true for every disruptive industry now challenged by an LLM, including my own older µSaaS product. Because I know that standing on solid ground without seeing the lava beneath is how even empires can fall.

Now, consider my neighbor again. Once I’d given him all of his options with mapped costs and risks, especially his personal risks — what option will he choose? Is this really the question about technology or is it about the customer? Who are the process-mining customers? Are they traditional companies or innovators? What is their mindset? Can the "little people" figure this out on their own? And what is really required here to come up with a durable answer?

Let’s fast-forward — the durable answer can only be produced by a human because "discovery" is necessary. But this does not diminish the role of the machine. Once discovered, pattern matching helps. The machine is there to still only augment the human. When this augmentation happens, and the right answer is produced, a tectonic shift occurs. Because a process was just created!

And this is the best hint we have yet. Doesn’t process-mining begin with discovery? Exactly where the time gets sunk and enhancements are being researched today?

The final question to my neighbor: "if you do discovery anyway — why do you wait?"

What is the Solid Ground?

Remember what I started this analysis with in 2023? The Emerging Business Identities and Domains. Augmented Thinkers — workers that were not there before. I built the phases of AI Adoption around changes in business due to human augmentation. We’re business. And such tools which operate directly on us to change us, also change the way we do business. This is the fundamental effect AI Adoption has on us, the humans.

Now consider: what if the top customer — say Pfizer — happily saving money applying process analytics to their legacy processes — decides to also attack from the other direction? Apply AI-Augmented workers directly to process design and improvement in-place. Analytics done or not — irrelevant.

And it’s not "will." It "has." By financial gravity this tends to first happen to the areas of business that need changes the most. And this is exactly the market segment process-mining companies market to.

I hope the answer is obvious now — move with the market.

The Doomed Mindset

The mindset of "the new tech is a competitor, and a weak enemy" is doomed from the start.

This is not a competitor, this is a category shift, because it shifts humans that mind the process — as we’ve already elaborated. How does one defend process mining against something that makes process mining optional? So, lift-and-drop will not work for a business model that is self-deprecating.

The Correct Move

We look broader, far enough to see the floating island move and move laterally.

I argue that hedging alone is not marginal because it bets on the status quo in times of change. Sure, there will always be an area of the market that is slow to change. Because the change is inevitable and the pace of change is speeding up, the status quo segment shrinks faster too. The viable move is aggressive expansion into the preemptive category.

Not "process mining enhanced with AI." Not "AI that needs process mining." Something else entirely — into the 80% where the movement is happening now. And solid footing on the old plateau can serve as a springboard while it lasts. This way my neighbor gets his LLM as part of a traditional and safe bet he won’t question.

---

Case Study: Generational Shift on Tectonic Shifts

I have a framework called MATILDA. Built it around 2000 for MLOps — before "MLOps" was a word. The µSaaS I mentioned before. It consolidates entire business systems into a trunk-based, bare-metal-OpSec distributed architecture — think Google Borg, early Airbnb. Deployed in a few places. Solves real problems for developers working with AI pipelines. From Expert Systems in 2005 DoD to modern day "stick agent here" stuff.

When I pitched it to investors over the years, I watched something interesting. And perplexing.

The 60-year-olds saw "another infrastructure play." Middleware. Plumbing. They’d seen middleware before. They knew what it was worth. They evaluated accordingly. The rejection I’d get: "elegance and pragmatism are too expensive for us."

The 45-year-olds saw "ML tooling." Hot space. Good timing. They evaluated based on comparable exits in the MLOps category. The rejection I’d hear: "niche, bare-metal only, enforces philosophy."

The 30-year-olds saw something else entirely. They asked different questions. Many questions. About integration. About dynamic workflows. About how developers actually think. They weren’t evaluating a product. They were imagining a relationship with it. They caught on to the elegance and cleanliness of design. And kept tugging on it, breaking my "value" pitch to the older execs.

If 30-year-olds were enterprise decision makers, I’d be a gazillionaire.

Same framework. Same pitch. Different generations. Different perceptions. Different conclusions.

The Living Laboratory

Now watch my own family. Our home is heavily augmented with technology. A lot of our own making.

I’m 52. Over three decades of hands-on designing and coding enterprise technology. I’ve seen mainframe to cloud to GPU and loved every stage. I recognize the patterns. I know what prime mover shifts look like. I architect based on what I’ve witnessed before. And I design around people.

My son is 16. Homeschooled for personal leadership development, so he has a say in his curriculum. He once happily made Minecraft mods in Java. Ran kids' coding communities. Then worked with me on real production systems — edge devices in Kotlin and Python.

He chose psychology and neuroscience as his major subjects this year. Not computer science. Not engineering.

When I asked why, he said: "The interesting problems aren’t in the code anymore. They’re in how minds work — all kinds of minds."

Now he’s using multiagent synthetic teams to study something other than engineering. He’s not interested in ML anymore. He’s learning to understand intelligence itself. He sees something I can only partially see. His native perception is different from mine.

My daughter is 3. She talks to AI assistants like she talks to family. No ceremony. No "wow, technology." Just conversation. She’s crystal clear about which assistant can do what. And she’s well aware that her 50-year-old nanny "doesn’t like our robots."

She will never know work without synthetic collaboration. The concepts I had to learn and accept, she absorbs like gravity — invisible, obvious, always there.

All the while my visiting mother asks: "What do you mean you can’t turn it off?"

And my daughter responds: "Baba, it’s like your iPad, but good, and you don’t need anything."

That phrase — "don’t need anything" — reveals the generation shift. Baba (grandma) gets around with trinkets: iPad, smartwatch, gadgets in tow. But to this toddler, her entire world is the trinket. Nothing more is needed to interact.

What This Means

Four generations. Four completely different relationships with the same shift.

I architect from pattern recognition. I see the steam engine and know railroads follow. But I can’t feel what my son feels. I learned this world. He was born into it.

My son studies the substrate itself. Not "how to use AI" but "how minds work." He’s preparing for something I can barely imagine. But he can’t see what I see — the historical patterns, the failure modes, the corporate gravity.

My daughter will build on foundations neither of us can predict. The concepts we struggle to explain will be her common sense.

My mother fights the glass rectangle daily. She sees threat where we see opportunity. Complexity where they see simplicity.

The Enterprise Implication

Your 50-year-old architects see patterns. They know what "can’t work" based on lived experience. Trust them. They’re often right. But they’re also blind in specific ways.

Your 28-year-olds feel the culture. They know what adoption looks like from inside. They have intuitions your 50-year-olds cannot have. But they lack historical depth.

Your interns and new graduates? They’re not just cheap labor. They’re the medium. They show you what "native" looks like. Watch how they relate to synthetic collaborators. That’s your future.

The winning move isn’t choosing one generation over another. And notice: each generation defaults to a different Pleb perspective. Your 60-year-olds think like Capitalists — ROI, exits, market position. Your 30-year-olds think like Engineers — how does it actually work? Your interns are Democrats — they just want the benefits, and they’re right to expect them. It’s building teams where each generation’s perception complements the others' blindness.

The 50-year-old sets policy the 28-year-old can’t yet understand. The 28-year-old embodies culture the 50-year-old can’t feel. The 16-year-old studies fundamentals neither thinks to question. The 3-year-old absorbs concepts both generations struggle to accept.

This is how you build for a tectonic shift. Not by predicting the future. By assembling the perceptions that can see it from all angles.

And this is what I tell CTOs now:

Design your company for your interns.

The tectonic shift has happened. Everything will spin faster and faster now. These youngsters are the substrate you’re building your business on. Your architects should target them.

---

Case Study: Origami AI Adoption — New Within Old

While American CTOs chase the latest AI vendor, a manufacturer I worked with years ago just achieved something remarkable.

They’re crushing AI integration — Phase III, production optimization, revenue climbing — using architecture we designed in 2018.

Seven years old. "Boring." And it works.

The Setup

Let’s call them "Origami Corporation." Specialized components for heavy machinery. Hydraulics, not hashtags.

Key details: - ~200 engineers total (mostly hardware) - Only ~30 software engineers - Part of a larger conglomerate - Extreme focus on production efficiency

In 2016, a Big American Software company (let’s call them "EvilCorp") promised to "transform" their business. By 2017, everything was worse. Costs tripling.

Their CTO called me in 2018.

The Architecture That’s Still Winning

We built something radical for 2018: a modular monolith with military-grade isolation.

The secret: "Revolving Door" security gateways.

Production calls out — never receives calls. Data enters the gateway. Door rotates (physical air gap, 4-hour cycle). Business systems receive cleaned data.

Paranoid architecture that accidentally created perfect AI boundaries.

Why "Old" Beats "New"

Here’s what American enterprises don’t understand: AI needs boundaries, not microservices.

Their AI integration timeline:

Phase	What They Did	Result
Phase I (2024)	AI at domain boundaries using existing gateways	Liability costs: -17%
Phase II (2024-25)	Cross-domain AI via existing ETL pipelines	Liability costs: -19%
Phase III (2025)	AI-driven production optimization	Liability: -23%, Revenue: +3%
Phase IV (now)	Autonomous coordination worldwide	In progress

Note the overlap — they run phases in parallel because their architecture allows it.

When Model Context Protocol emerged, it validated everything. Simple context boundaries. Clear data contracts. Domain isolation.

They implemented MCP in two weeks. How long would it take you?

The Lesson

Your 500 microservices can’t do this. Their 4 modules can.

Not because of implementation choices — because of clean architecture.

While you were deploying Kubernetes, they kept their modular monolith. They perfected their boundaries. They focused on their business.

Boring architectures print money. Exciting architectures print resumes. This is what Engineering Collective discipline looks like. No Capitalist hype-chasing. No Democrat trend-following. Just substrate. Just boundaries. Just work.

The Best Part

They haven’t called since 2019. Just reached out to tell me how AI is going.

That’s the point of good architecture. It survives its creator.

Seven years later, they’re implementing advanced AI without consultants. Meanwhile, your competitors hire me every 18 months to fix the same problems.

If your architecture isn’t this clean, the path is harder but not impossible. Start with boundaries. Even one clean domain is better than zero.

Which company would you rather be?

For the Reader

If a "boring" manufacturer can implement Phase III AI with 30 engineers and 7-year-old architecture — what’s your excuse?

• Too complex? They manufacture for aerospace.

• Too regulated? They deal with military contracts.

• Too legacy? They still run Oracle RAC at the core.

• Too small? They have 1/10th your IT staff.

The difference isn’t capability. It’s discipline.

The companies winning with AI aren’t the ones with the newest tech. They’re the ones with the clearest boundaries.

Stop chasing magic. Start building foundations.

---

The Complete Argument

THESIS: AI is categorically different. It changes what humans ARE, not what they DO. The feedback loop now includes human cognition as a variable.

PROBLEM 1 - CATEGORY BLINDNESS (Celonis): Even smart companies standing on solid ground can miss the lava beneath. They see competition when it’s category shift. They defend their position when the entire paradigm is moving. The danger isn’t losing to competitors — it’s the ground itself floating away.

PROBLEM 2 - PERCEPTION BLINDNESS (Generational Shift): Different generations see the same shift completely differently. The 60-year-old sees plumbing. The 30-year-old imagines relationship. The 3-year-old doesn’t need anything to interact. No single generation can see the whole picture. Each is blind where others see clearly.

THE SECRET: The technology isn’t the model. The technology is the relationship. Team B (collaborator mindset) gets different outcomes than Team A (tool mindset). Same compute, different emergence.

PROOF - WHAT ACTUALLY WORKS (Origami): A "boring" manufacturer with 30 engineers and 7-year-old architecture is crushing Phase III AI. Not because of new tech — because of clean boundaries. Discipline beats hype. Foundations beat trends. Clarity needs silence!

THE PATH FORWARD:

1. Recognize the categorical shift (not incremental);

2. Assemble multi-generational perception (no one sees everything);

3. Build relationship, not tooling (Team B mindset);

4. Create boundaries, not complexity (architecture discipline);

5. Design for your interns (they’re the substrate now).

The 2026-2028 Window

This is the year of no return.

I’ve spent 2025 talking to enterprises about this shift.

Out of every hundred decision-makers, five to seven understand what’s happening. Out of those, one is actually building relationships with synthetic collaborators.

The rest are waiting. For maturity. For consensus. For someone else to prove it works.

That wait is the drift.

Interestingly, the spread is the same among senior developers and architects. Junior developers and "talkers" don’t factor in — there’s no point-and-click trivial tool to run. My continuous deep search finds that among influencers, ONLY JetBrains speaks truthfully to these numbers: Top Developer Trends of 2025: AI Usage; Productivity; and the Rise of TypeScript. (And Ciara Byrne (LI) of JetBrains is incredible — I deeply respect her work.) Everyone else overhypes the latest LLM achievements even when they’ve never run any.

This drift is not dramatic. Not visible. Just…​ quiet separation.

• By 2027, the gap will be measurable.

• By 2028, the ones who waited will wonder what happened.

This does not mean laggards go bankrupt immediately. It means two classes of companies with a chasm between them. A "commodity class" with hard ceilings on margins, paychecks, and profits — that’s the 95%. And a ruling class of technology-first companies: great compensation, deeply technical interviews, tool literacy required.

The <5% won’t explain. They’ll be too busy building the future.

What This Means for the Three-Plebs

Some winning and some losing is not the sad part here — that is natural and necessary.

The sad part is what it does to our Three-Plebs-Collective.

The Capitalists who bet right will thrive — they always do. The Democrats will follow whoever wins — they always have.

But the Engineers — the middle class, the builders — they’re sized to the market. When only <5% of companies cross the divide, the market for skilled technical work shrinks with it. This is exactly the formula we’ve watched repeat since the dot-com bubble. Each disruption concentrates.

The only way to grow the middle class is to grow the economy. And the only way to grow the economy now is broader AI adoption. More augmented contributors means more GDP growth.

Here’s the uncomfortable truth: The automobile didn’t make cars expensive. It made horses cheap. The automobile created more jobs than it destroyed — but only because adoption was universal. If only 5% of people drove cars, the economy would have shrunk, not grown. Personal AI augmentation does the same thing to un-augmented humans.

The question isn’t whether this happens. It’s whether you’re the car or the horse.

---

This article is for the ones who sense the ground moving. You don’t need to be certain. You just need to start.

The islands are already drifting apart.

Which one are you standing on?