Pull Not Push: Why We've got AI's Impact on the Future of Work All Wrong

I have spent roughly 30 years in the construction and property industry. Over that time, I have worked across the messy reality of projects, budgets, documentation, consultants, contractors, clients and all the fragmented information that sits between a decision and an outcome. For many years I have also been trying to build a technology platform around that reality. For a long time, it felt beyond reach. The ambition was clear, but the challenge and the cost was too big. Now, for the first time, it feels genuinely possible.

But the opportunity I see runs counter to what I think most people think AI will contribute - a team of agents with minimal humans. I don't think this is the case and that perhaps the current discussion about AI is pointed in the wrong direction. The loudest narrative is still about replacement: replacing jobs, replacing knowledge workers, replacing professional judgment. In some industries that may be part of the story, but in most businesses the bigger shift is not replacement. It is reconfiguration. AI has the opportunity to make it possible to redesign how information is gathered, how decisions are prompted, how knowledge is surfaced and how work is reviewed.

As noted from the outset, my own perspective comes from construction and property - an industry synonymous with productivity losses over time where other industries have had substantial gains - but the underlying point is broader than that. Most organisations do not suffer from a total absence of information. They suffer from fragmented information, delayed information, unstructured information and inconsistent follow-through. They suffer from systems that depend too heavily on people manually pushing or extracting data so it can end up at the right place at the right time. That is where the future of work is changing. For the Construction, Architecture and Property Industry it change cannot come quickly enough. Here are some thoughts that occurred while I was reframing what the productivity landscape looks like in 2026 and beyond.

1. Pull, Not Push, Documents and Information

The first shift is from users pushing documents into the system to the system pulling them from wherever they already live. Outlook, WhatsApp, spreadsheets, supplier emails, bank feeds, credit card data and other business systems should be treated as source environments, not separate admin tasks.

Pull not Push solves a problem that exists in almost every industry: poor data input from the field. The issue is rarely that the information does not exist. The issue is that nobody reliably uploads it, categorises it and keeps it current.

Most businesses are still built around the fantasy that disciplined manual input will solve the problem. It usually does not. For someone on a scaffold installing windows forwarding information to head office is problem number 100 in the context of their primary role and the things that could go wrong on any particular day.

The better model is to automate ingestion as much as possible, then let people review, approve and intervene where needed.

That is a very different idea of work. Instead of relying on human discipline to collect and move information around, the system becomes responsible for gathering the raw material of decision-making. People can then focus on what matters more.

2. Pull, Not Push, Prompting of People

The second shift is the same principle applied to workflow. The product should not wait for the user to ask questions. The agent should prompt the user.

That means an AI layer that uses retrieval, classification and pattern detection to ask the right strategic question at the right time. What needs attention today? What is missing? What is late? What has changed? What requires a decision? What has drifted out of range? The future product is not mainly an input tool. It is an agentic decision-support system.

This is where a lot of current software still feels old. It waits passively for someone to log in, search, interpret, remember and act. But a good operator in any business does not work like that. Experienced people carry an internal map of what needs watching. They know which signals matter and which questions to ask. AI becomes useful when it starts to replicate some of that prompting function and place better questions in front of people at the right time.

3. Data Mapping That Creates Context

The third shift is to stop treating documents as isolated files. Systems should map relationships between documents, events and financial records so the user can understand context.

An RFQ should connect to a quote, which connects to a purchase decision, invoice, committed cost and cost-to-date reporting. Different versions of the same document should be linked. The user should be able to see the chain of information through a graph, timeline or dependency map. That is how documentation becomes legible rather than fragmented.

This matters far beyond construction. In almost any business, people are dealing with chains of information rather than single documents. Emails relate to contracts. Contracts relate to approvals. Approvals relate to invoices. Invoices relate to performance. What most systems still do badly is help people see those relationships clearly. AI can help make those links visible and usable.

4. Multiple Ways to Analyse a Project or Business Problem

Complex work should not be analysed through a single lens. The system should build a contextual position using multiple models at once.

In my world that means combining analogous project benchmarking, square metre rates, granular quantities, trade-specific measures, programme logic, labour hours, durations and delivery history. In another business, the equivalent might be customer cohorts, margin trends, operational throughput, contract value, staff utilisation and delivery timing. The point is the same. Work needs to be assessed through both top-down and bottom-up models, not through one narrow method in isolation.

That is a more evidence-based approach to management and finance. A single pricing method, forecasting method or reporting view rarely tells the whole story. Better systems should be able to hold multiple forms of analysis in tension and help the user arrive at a more grounded position.

5. Data Sets Creating a Feedback Loop

Once the system is consistently ingesting and structuring data, it can create a real feedback loop. That is where the product becomes genuinely intelligent.

Historic and live data can then be used to validate or invalidate a position. Is this budget realistic? Is this programme credible? Is this subcontractor cost in range? Is this team capacity assumption plausible? Is this project or workstream tracking in line with similar jobs? The point is not just to store information, but to create a growing evidence base that improves future forecasting, financial control and decision quality.

This is one of the most under-appreciated changes in the future of work. For years, businesses have collected data without really learning from it. AI creates the possibility of a live feedback system where prior decisions, current conditions and emerging outcomes are constantly being compared. That is when experience becomes more scalable and forecasting becomes less speculative.

6. Human Review, Not Manual Administration

A further point is that the goal should not be to remove human judgment. It should be to remove low-value admin.

Most of the work around collecting information, reconciling it, categorising it and linking it is rule-based and algorithmic. That should be automated. The human role should shift toward review, approval, exception handling and strategic judgment.

I think this is the part that both AI evangelists and AI sceptics often miss. The best use of these tools is not usually to eliminate the professional. It is to reduce the amount of energy highly capable people spend on mechanical tasks that software should have taken care of years ago. The future of work is not less judgment. It is more judgment applied at a higher level of leverage.

7. Information Architecture and Provenance

If a system is going to pull from multiple channels, it also needs a clear information architecture. Users need confidence in where the data came from, which version is current and how one item relates to another.

That means provenance, version control and traceability are not side issues. They are core to trust in the platform. If people do not trust the chain of information, they will always fall back to manual checking, parallel spreadsheets and side conversations. In other words, they will revert to old ways of working.

The future of work will depend not only on better automation, but on better trust. Systems will need to show their workings. They will need to make source, sequence and relevance visible. Without that, AI remains interesting but not operational.

The Real Shift

From where I stand, after three decades in industry and many years trying to build technology that could genuinely support it, the most important shift is this: the future of work is moving from manual administration toward structured review, from isolated files toward connected context, and from passive software toward active prompting.

That is the counterpoint to the mainstream story. The future is not mainly about AI replacing people. It is about AI making it possible to redesign work around better information flows, better feedback loops and better use of human judgment.

That is why this moment feels different to me. For years, the vision of a system that could pull fragmented information together, map it, question it and support better decisions felt out of reach. Now it does not. The tools have finally caught up with the problem.

And that may turn out to be the real story of AI at work: not that it removes the need for experienced people, but that it finally gives them a system worthy of their experience. a perennial problem is lack of information from the field, for someone on a scaffold installing windows forwarding information to head office is problem number 100 in the context of their primary role and the things that could go wrong on any particular day.