Five Frontier AI companies. Fourteen days. The same answer.

March 9: @Microsoft launches Agent 365, a control plane for AI agents embedded across every workflow in the enterprise. Not a product. Infrastructure.

March 11: Perplexity®️ ships Personal Computer. An always-on agent that lives on your Mac Mini, runs 24 hours a day, and connects to everything you work with.

March 16: Meta puts Manus on your desktop. Local files, local apps, multi-step tasks. No prompt required.

March 17: Google releases Gemini Computer Use and simultaneously deploys Google-Agent, a new web-level identifier for AI agents acting on users’ behalf. Not a device. A protocol. Google is not competing for residency on your machine. It is competing for residency on the web itself.

March 23: Anthropic gives Claude computer use and Dispatch. An agent that completes tasks on your machine while you are away.

Three weeks. Five of the most capitalized technology companies in the world. All converging on the same architecture.

The answer they converged on: AI needs a home. Not a browser tab you open and close. A persistent system with access to your files, your calendar, your inbox, your decisions. One that runs whether you are watching or not.

They are competing on residency now. Who owns the environment the agent lives in. Who controls the persistent context, the local files, the background processes, the connection to your calendar, your inbox, your decisions.

That is a land grab for the layer that sits between the model and your life.

When companies with this much capital start racing on proximity, it tells you something precise.

Raw intelligence is no longer the scarce resource.

Access is.

Most people described what happened as a product race.

New tools, new capabilities, new announcements. The usual cycle of press releases and benchmark scores.

That framing missed the actual shift.

This is not a product race. The labs that win it do not just have the best model. They have the ambient system you cannot operate without. The one running in the background before you open a browser, before you start a meeting, before you make a decision.

That is a different kind of asset than a product.

That is infrastructure.

When AI systems could only respond to prompts, they were tools.

Sophisticated, impressive, occasionally transformative. But tools. You picked them up, used them, put them down.

The agent layer changes the fundamental relationship between software and work.

An agent does not wait to be prompted. It holds context, executes across sessions, monitors for conditions, and initiates action.

It does not answer questions. It runs processes.

The moment software can run a process end to end without a human initiating each step, the entire category of work that required a human to be present collapses in scope.

That is not a productivity improvement. That is a structural change in what software can own.

The industries with the most exposure are not the ones getting the coverage.

Healthcare administration. Not clinical care. The operational layer. Prior authorizations, billing disputes, care coordination, appointment routing. A 2023 analysis of CMS hospital cost data by Trilliant Health found that administrative costs at U.S. hospitals reached $687 billion, compared to $346 billion in direct patient care. Nearly two dollars spent on administration for every dollar spent on treating a patient. These processes exist not because they add clinical value but because they require coordination across incompatible systems. Agents trained on those systems navigate them faster, more accurately, and without the variability of human fatigue.

Legal services. Not courtroom advocacy. The research, drafting, contract review, due diligence, and compliance work that drives the majority of billable hours at mid-market firms. Recent benchmarks cited by Fennemore Craig, a law firm that published its own analysis of AI adoption, show AI-enabled associates completing NDA drafts up to 70 percent faster than non-AI peers. Clients have already started demanding what the legal industry is calling AI discounts in 2026 panel reviews. The billable hour model is repricing whether firms are ready or not.

Physical supply chains. The scheduling, routing, exception handling, and vendor communication that keeps goods moving. None of this required creativity. It required consistency, scale, and the ability to process status updates across dozens of systems simultaneously. That is exactly the task architecture agents are designed for.

Financial services. Not trading. The compliance review, audit preparation, risk reporting, and client documentation that consumes the majority of hours at every mid-market firm. The work that was never glamorous enough to automate first, and never simple enough to automate easily. It is both now.

None of these industries described themselves as software businesses. They all are now.

Nobody wants to say this part directly.

Your expertise and institutional knowledge are no longer your moat.

They never were. They just felt like one because they were expensive and slow to replicate.

One lawyer with the right agent stack closes deals, reviews contracts, and runs due diligence at the output of a ten-person team. One analyst with the right infrastructure models risk, prepares reports, and surfaces anomalies faster than a department. One engineer directing a fleet of coding agents ships at a velocity that would have required a full org chart two years ago.

The expert with leverage is not replacing the expert without it.

The expert with leverage is replacing the team the other expert used to need.

Which means everyone whose income depends on expertise alone, without an ownership layer above it, is not competing against AI. They are competing against a version of themselves who figured this out earlier.

Your leverage, your distribution, your capital base. All of it needs to be rethought against that reality.

When production gets cheap, what gets expensive?

Taste. Knowing what is worth building before everyone else does.

Judgment. Knowing when to stop, when to double down, when the market has moved.

Trust. Getting other humans to act on your output, not just consume it.

Distribution. Reaching the right people before the noise does.

Capital. Compounding gains from one cycle into the next without starting from zero.

None of these live in a codebase. All of them have been underinvested in because technical skill was the faster path to income.

That trade-off no longer holds.

When the cost of production collapses, the value shifts entirely to the layer above production.

It happened with music. Recording democratized. Distribution democratized.

The artists who endured were the ones who owned the relationship with the audience, not just the ability to record.

It happened with publishing. Printing democratized. Distribution democratized. The writers who won had a point of view so distinct that readers sought them out directly.

Now it is happening across every industry that runs on credentialed human judgment.

The moat is what you build above the expertise. The systems you own. The trust you have built at scale. The capital that compounds independent of how many hours you log.

Most people have spent their entire career building the expertise.

Almost no one has built the layer above it.

A new kind of operator is appearing and nobody has named them properly.

The people who learn to direct agents effectively stop being individual contributors. They stop being managers too. They become systems operators. Setting objectives. Evaluating outputs. Catching drift before it compounds. Building loops that get smarter with every cycle.

It is closer to what an architect does. Or an investor. Someone who designs the system rather than runs inside it.

The role exists. The title does not. That gap will close fast.

Jobs will be lost. That part is not a debate anymore.

Every wave of automation displaced workers. This one is faster and reaches further up the skill ladder than any before it. White collar. Knowledge work. Code. The roles people spent decades building credentials for.

That is happening. Accept it.

But fixating on displacement is the wrong obsession. It is like watching a flood and arguing about which houses get wet first, when the real question is who thought to build on higher ground.

Job loss is a symptom. Not the disease.

It is about who understood early enough that expertise without ownership is just a job with extra steps.

The companies that survive this transition are not the ones adding AI tools to existing workflows. They are the ones rebuilding their operating models from the agent layer up. One trims the edges. The other changes what the business costs to run.

The incumbents who survive will be the ones who recognize this is not an upgrade cycle. It is a rebuild.

The companies that understand this earliest are not deploying AI. They are redesigning their org chart around what agents can own permanently, and building human roles exclusively around what requires judgment, accountability, and trust at scale. The question every leadership team should be answering right now is not how many people AI can replace. It is what operating model becomes possible when the cost of execution drops to near zero, and whether they are building toward that model or defending the one they already have. Anthropic shipped 74 product releases in 52 days. That is not a company adding AI features. That is a company rebuilding from the agent layer up, in real time.

The individuals who understand this earliest are not optimizing their skills. They are building something that compounds after the work stops. Right now, every hour they log pays them once and disappears. Every high earner whose income depends entirely on showing up is one restructuring away from zero. The agent fleet does not change that math directly. It changes the leverage available to anyone willing to build above the expertise line, which means the gap between those who do and those who do not will widen faster than any prior technology cycle.

So here is where this goes.

By the end of 2026, multiple industries besides technology will publicly cite agent-based automation as the primary driver of structural workforce reductions in knowledge work roles.

Not manufacturing. Not customer service. Knowledge work. Legal research. Compliance review. Healthcare administration. Financial analysis.

The announcements will be framed as operational efficiency.

The actual driver will be that one person running a fleet now does what ten people did before, and the economics of maintaining the ten are no longer defensible.

When those announcements come, the conversation will shift from “AI is coming for jobs” to “AI already took them while we were debating whether it would.”

The industries most affected will not be the ones anyone predicted.

They will be the ones that thought their complexity was a moat.

Complexity is not a moat against software. Complexity is what software is built to consume.

Marc Andreessen was right in 2011.

He was also early by fifteen years.

Software spent the 2010s eating the industries that were already information businesses. Media, retail, finance at the consumer layer, logistics at the interface layer.

2026 is different. Software is eating the judgment inside every industry. The kind that took decades to certify and minutes to route around.

The human no longer has to be in the room.

The software is already there. Running in the background. Holding context. Waiting for the next process to initiate.

Everything is now software.

Not the tools you use. Not the products you buy.

The operating layer of the entire economy.

The people who will feel it first are not the ones with the least expertise.

They are the ones who built everything on expertise alone and assumed that was enough.

Leave you with two questions.

If one person with an agent fleet can replace your entire team, what exactly are you selling that they cannot?

When the apprenticeship layer disappears and senior expertise becomes non-renewable, who trains the next generation of practitioners?

—

The author is co-founder of BricksFolios | Wealth-Tech for Tech Professionals, a real estate platform serving senior technology professionals.