Future signal
The public internet was only the starting point
The first wave of AI established one clear fact: the public internet was enough to get AI started, but it is not enough to make one business different from another.
Large AI models learned from articles, forums, documentation, books, code, videos, and almost every other kind of public information that could be collected at scale. That gave them language, general knowledge, and a broad understanding of how people communicate.
The public internet works like a shared instruction manual. It teaches the machine the general rules, but it cannot show the machine how your business actually works when a customer complains, a payment is late, or an important decision does not fit the usual process.
That is the shift businesses need to understand.
The next advantage will not simply come from having access to AI. It will come from giving AI the right business context.
The Public Internet Gave Everyone the Same Starting Point
Most companies now use AI models trained on broadly similar public information. The tools may differ, but the general knowledge inside them often comes from the same starting point.
An AI model can write an email, summarise a document, generate code, or explain a common business idea because examples of that work already exist online. What it cannot automatically understand is the judgement inside your business.
It does not know:
- How your best salesperson handles a difficult objection.
- How your support team resolves an unusual customer problem.
- How your operations team works around a broken process.
- Why one manager approves something and another rejects it.
- How your company balances speed, risk, margin, quality, and compliance.
- Which small exceptions and shortcuts keep the business moving.
That knowledge is rarely written down in one clean document. It is spread across employee behaviour, internal systems, meetings, ticket histories, edits, reviews, approvals, and decisions.
The important point is this: access to the model is becoming common. Clear business context is not.
Work Is Becoming Training Data
Meta provides one of the clearest examples of this shift in practice.
In April 2026, BBC and CNBC reported that Meta was introducing a tool called the Model Capability Initiative. The system was designed to capture employee activity such as clicks, keystrokes, mouse movements, and screen context to help train AI agents. CNBC reported that it could observe activity across tools and websites including Google, LinkedIn, Wikipedia, GitHub, Slack, and others.
The reasoning was straightforward. If an AI agent is expected to perform everyday computer tasks, it needs examples of how people actually use computers.
Text alone is not enough. The system needs to see the order of the work: what someone opens, what they click, what they copy, what they ignore, what they correct, and when they decide to escalate.
In June 2026, BBC reported that Meta scaled back parts of the plan after employee backlash. Workers were given controls that allowed them to pause collection for short periods and request exemptions.
The lesson is clear: the opportunity and the risk arrive together. The same activity can look like valuable training material to a business and job-copying surveillance to an employee.
Both reactions make sense.
A Softer Version Is Already Mainstream
Meta is the sharpest example, but the wider idea is already becoming normal.
Microsoft's 2026 Work Trend Index describes companies becoming "Learning Systems" that capture signals from work and use them to improve how the organisation operates. Microsoft says it analysed anonymised Microsoft 365 productivity signals, Copilot usage, and agent activity to understand how work is changing.
This is a more polished version of the same principle.
A business is no longer only a place where work happens. It can become a system that learns from the work being done.
Every document, prompt, workflow, customer interaction, correction, handoff, and approval can add to the company's intelligence.
The real advantage comes from how well a business captures that knowledge, organises it, protects it, and uses it again. Collecting more data is not the same as creating better understanding.
The Opportunity Is Hidden in Everyday Work
The most valuable business knowledge is not the document, ticket, or spreadsheet itself. It is the judgement behind it: why someone chose one response, rejected another, made an exception, or escalated a decision.
Support tickets reveal why a resolution worked. Sales conversations show which objections matter. Finance approvals expose risk thresholds, while operations and compliance records capture recurring problems and important exceptions. AI can make these patterns easier to find and apply, helping teams learn faster and make more consistent decisions.
The goal is not to collect everything or replace expertise. Capture selected, high-value knowledge, record its context, exclude sensitive information, and assign someone to review how it is used. This turns individual judgement into a reusable business capability.
The Risk Is Losing Trust
The obvious risk is that companies treat this only as an opportunity to watch people.
If employees believe their work is being recorded so that their jobs can be copied, trust falls. That pressure changes behaviour: people avoid experimenting, hide mistakes, and resist the AI system.
A machine can only learn from the inputs it receives. If fear changes the way people work, the business collects more data while learning less about how good work actually happens.
There are also serious privacy, security, legal, and ethical questions. Internal work data often contains customer information, employee personal data, passwords, health details, confidential strategy, regulated material, or commercially sensitive information.
A business cannot record everything first and decide what was acceptable later.
It needs to separate:
- Data used to improve a workflow.
- Data used to evaluate an employee.
- Data used to train an AI system.
- Data that should never be captured.
Confidentiality is not just a sentence in a policy. It has to affect how the data is collected, who can access it, how long it is kept, and what the system is allowed to learn from it.
What Businesses Need to Do Now
Do not start by monitoring everything. Start by identifying where valuable judgement already exists.
Ask:
- Which workflows contain the most valuable human judgement?
- Which tasks happen often enough to be worth modelling?
- Where are people making decisions that have never been documented?
- Which internal data sources can safely improve AI performance?
- What information should be excluded for privacy, legal, security, or trust reasons?
- How will employees be told, involved, and protected?
- Who owns the knowledge created from this work: the team, the company, or the technology provider?
Then choose one workflow and record it properly. Document the steps, the decisions, the exceptions, a few good examples, and the points where a person still needs to check the work.
A plan makes execution easier. The same applies here. Once the workflow is clear, the business can make a better decision about what AI should support, what it should never do, and what evidence would show that it is actually helping.
The Bottom Line
The public internet gave AI its general knowledge. The next layer will come from private, business-specific information: the real patterns of how work happens inside a company.
That creates an opportunity, but it also creates a responsibility.
The future workplace will not only use AI. It will teach AI.
The businesses that handle this well will not be the ones that collect the most data. They will be the ones that know what they are trying to learn, protect the people and information involved, and turn everyday work into useful knowledge without turning the workplace into a surveillance system.
The standard is simple: capture with a purpose, protect by default, and keep human judgement accountable for what happens next.