The attention thesis
Follow the Human Attention
If you want to understand why Salesforce, Adobe, Shopify, and the software economy are under pressure, stop following the screens. Follow where human attention is being spent.
Predicted
Actual
Difference = surprise
Start with the flag
If you want to understand why companies such as Salesforce, Adobe, and Shopify are under pressure, look at where human attention is being spent.
If you see a salesperson sitting in front of a computer updating CRM records, imagine a referee throwing a flag.
A salesperson’s attention belongs in relationship building, negotiation, judgment, trust, problem solving, timing, and finesse. The CRM record may be important, but importance does not determine what deserves attention.
Your heartbeat is important. Breathing is important. Digestion is important. Yet you do not ordinarily attend to them.
The subconscious does not beat your heart. Your heart beats.
The subconscious does not breathe. Your respiratory system breathes.
The subconscious predicts these activities. When the predicted heartbeat and the actual heartbeat match, there is no surprise demanding your attention. When prediction and actuality separate, attention returns.
That is the essential distinction.
AI is not first a worker. It is not first an agent. It is not fundamentally another collection of automated workflows.
AI is first a prediction machine.And that is why conventional business software is under threat.
01
Software stores.
AI predicts.
Traditional software is built around databases.
Salesforce stores customers, contacts, opportunities, activities, forecasts, and project records. Shopify stores products, customers, orders, inventory, and transactions. Adobe stores documents, images, designs, layers, edits, and media assets.
The software database contains records of what the organization believes to be actual.
That database then requires an interface. Humans must create records, update fields, attach files, move opportunities through stages, correct errors, run reports, and maintain the digital representation of the business.
AI does not need to reproduce that structure.
AI does not need its own CRM database. It predicts the CRM artifact.
It does not need to store the actual CAD drawing. It indexes the CAD drawing where it exists as a source of truth.
It does not need to store the transcript from a meeting. It indexes the actual transcript.
It does not need to create another permanent copy of the customer order, contract, article, permit, specification, or invoice. It indexes the source of truth and uses those actual artifacts to improve its predictions.
Software accumulates records.
AI accumulates predictive capacity.
That difference changes everything.
02
The CRM is a monument to misplaced attention.
Management has a familiar saying:
If it isn’t in the CRM, it doesn’t exist.
But the opportunity does exist. The customer exists. The relationship exists. The building exists. The salesperson’s knowledge exists.
What management means is that the opportunity does not exist as a stored software artifact until a human represents it inside the CRM.
Someone must create the account. Add the contact. Enter the opportunity. Assign a stage. Estimate a probability. Attach the proposal. Record the next action. Explain why the bid was lost.
These records produce useful outputs. The CRM generates a forecast. It connects sales to accounting. It supports commission calculations and production planning. It gives management a dashboard.
But the entire system depends on human beings continually telling the database what is happening in Reality.
That is the weakness AI exposes.
03 / Case study
The Four Seasons opportunity.
Consider a subcontractor operating within fifty miles of Charleston, South Carolina.
In the conventional software model, the process begins when an invitation to bid arrives. Perhaps it appears in PlanHub. Perhaps a general contractor sends a request for proposal by email.
Someone reads the request and creates an opportunity in the CRM. The plans are attached. The project is categorized. An estimator is assigned. The deadline is entered. The opportunity moves through a series of stages until it is marked won or lost.
AI changes the beginning of the process.
It does not wait for the bid.
The prediction may begin with a newspaper article announcing that a Four Seasons hotel is coming to Charleston. The AI indexes that article as a source of truth. From there, it begins predicting the artifacts that do not yet exist.
It predicts the opportunity.
It predicts the likely contacts.
It predicts the developer, architect, general contractor, consultants, and decision-makers.
It predicts the project’s probable scope, specifications, square footage, budget, schedule, legal terms, insurance requirements, and competing subcontractors.
These predictions are not stored actualities. They are expectations about artifacts that Reality has not yet authored.
As permits, announcements, architectural documents, emails, meeting transcripts, and bid materials appear, the AI indexes each source of truth. The actual artifacts continuously test the predicted artifacts.
Where prediction and actuality match, no human attention is required.
Where they differ, surprise appears.
That difference tells the organization where human attention belongs.
The AI may predict that the appropriate human action is to call a particular person. Attend an event. Ask for an introduction. Visit the site. Review an unusual indemnification clause. Reconsider the estimated labor requirement. Speak with a supplier before material prices change.
Those are predictions about where human attention will produce the greatest value.
The human still acts.
The salesperson makes the call. The estimator exercises judgment. The executive accepts the risk. The attorney interprets the unusual term. The subcontractor builds the relationship.
The prediction machine predicts.
The human attends and acts.
04
PlanHub and the CRM arrive too late.
This is why the threat is larger than AI filling in CRM fields.
PlanHub delivers the opportunity when the bid has already been organized and distributed. The CRM records the opportunity after someone in the organization recognizes it.
The prediction machine begins earlier.
It can identify an emerging opportunity before the formal bid exists because it is predicting forward from sources of truth. By the time the exact plans and specifications arrive, those documents are not the beginning of the opportunity. They are actual artifacts against which an existing prediction can be compared.
The CAD drawing remains where the architect placed it.
The contract remains in the document system.
The meeting transcript remains with the meeting record.
The invoice remains in the accounting system.
The AI indexes these sources. It does not need to gather copies of everything into a second database and ask humans to maintain them.
This is what threatens the CRM’s position as the center of the organization.
The database may continue to exist. Software may still store the legal customer record, the executed contract, the official invoice, and the completed transaction. But humans no longer need to spend their attention operating the database.
05
From automation to absorption.
Traditional automation transfers execution. A workflow performs a predefined step.
Absorption transfers attention.
Absorption occurs when the predicted artifact and the actual artifact are the same. When the prediction of the project, contact, specification, billing code, forecast, contract, or next appropriate human action matches the actual artifact, there is no surprise.
No surprise means no demand for human attention.
This process will not necessarily look like a dramatic software migration. The organization may continue paying for Salesforce long after people have stopped using it in the traditional way.
- The screens become quieter.
- The records stop receiving manual updates.
- The dashboards become less relevant.
The prediction machine indexes the real sources and predicts the artifacts the CRM once required people to construct.
Then one day the CRM breaks, and nobody notices.
That is when absorption is complete.
06 / Incumbents
The infrastructure may survive while the software experience disappears.
Why Adobe is under threat
They do not want Photoshop. They want the image.
The same distinction applies to Adobe.
Photoshop is software. It stores and manipulates an actual image artifact. Its interface gives a human access to layers, masks, selections, brushes, corrections, and effects.
But most organizations do not fundamentally want someone to operate Photoshop. They want an image.
The prediction machine predicts that image.
It predicts the appropriate composition, lighting, background, dimensions, branding, product placement, and campaign variation. The predicted image can be compared with the image the organization actually accepts and uses.
When the predicted artifact is sufficiently close to the actual artifact, the steps inside Photoshop lose their claim on human attention.
This does not mean every form of photography or design disappears.
A wedding photographer is attending to a real event. Presence, trust, timing, human emotion, and physical perspective matter. A creative director may need to attend to taste, meaning, cultural context, and the final decision.
But masking an object twenty times, resizing the same image for twelve channels, or mechanically producing campaign variations is an inappropriate use of human attention.
Adobe can add AI throughout its products, and it is doing so. But that does not remove the strategic danger.
If the prediction becomes the valuable artifact, the tool used to manually construct the artifact is no longer the center.
Shopify and predicted commerce
The merchant should not spend the morning in control panels.
Shopify faces a similar transition.
Its software stores actual products, orders, payments, inventory, and customer records. Those actual artifacts still matter. Commerce requires authoritative records of what was ordered, what was paid, what was shipped, and what remains in inventory.
But a prediction machine does not need to turn Shopify into another place where humans continually maintain those records.
It can index the actual order, inventory, payment, product, and fulfillment sources. From those sources, it can predict demand, pricing, merchandising, promotions, creative assets, customer questions, purchasing behavior, inventory shortages, and the next appropriate action.
The merchant should not have to spend the morning moving between control panels.
The merchant should attend to the surprising customer problem, the unusual product opportunity, the supplier relationship, the brand decision, or the consequential risk that the prediction machine cannot absorb.
Shopify may remain valuable as transaction infrastructure. Payments, identity, compliance, and fulfillment still require actual records.
But the visible interface and its demand on human attention become less valuable.
The infrastructure may survive while the software experience disappears.
07
What the stock market is seeing.
This explains why strong current results can coexist with falling software valuations.
Salesforce and Adobe have continued reporting strong results. The market is not simply judging what these companies earned yesterday. It is attempting to price the future value of their software if AI absorbs the activities performed inside it. Salesforce’s results, Adobe’s results
The old software model monetized storage, interfaces, workflows, and human seats.
The prediction model monetizes the accuracy of predicted artifacts and the amount of human attention returned to the organization.
That creates a difficult problem for incumbent software companies. Their AI products may grow rapidly while simultaneously weakening the original reason customers paid for their software.
The better the prediction machine becomes, the fewer people need to operate the underlying application.
The fewer people who operate the application, the weaker the economics of charging by the seat.
Adding AI to every screen does not solve this problem. If the AI is successful, people should need fewer screens.