Somewhere right now, a customer is repeating themselves. They are explaining their problem for the third time, to the third person, because the organization on the other side has no shared memory of the previous two conversations. It is an infrastructure problem that AI is making harder to ignore.
It is also becoming impossible for policymakers to ignore. Just in April, the Mayor of London launched a new AI and Jobs Taskforce to examine how AI is changing work across the capital, signaling that the conversation has moved well beyond investment announcements and into the harder question of what AI does inside organizations.
It is also shining a spotlight on a memory crisis inside modern business.
AI is accelerating work, not clarity
As UK organizations rush to deploy AI in the workplace, many are layering it onto fragmented systems that were never designed to preserve institutional memory in the first place.
According to research published in Harvard Business Review, knowledge workers toggle between applications and tools roughly 1,200 times per day, a pattern known as “toggling tax”. That figure alone tells the story: we aren’t short of tools, but there is no coherence among them.
The result is a new kind of productivity paradox. Work is moving faster, but clarity is not improving.
This is where much of the current enterprise AI conversation unravels. A surprising amount of what is marketed as AI today still relies on humans to do the synthesis work themselves. The system retrieves documents. It summarizes conversations. It surfaces links. But employees still carry the burden of reconstructing meaning, and so do the customers and end-users waiting on the other side of those decisions.
Notably, when these types of AI tools do the retrieval, but humans skip the synthesis, the output feels hollow. That creates a trust and credibility problem – not just for the individual, but for AI as a category. People start associating “AI-assisted” with “low-effort”.
When context is lost internally, the effects aren’t invisible. They surface as slow responses, repeated requests for information that customers already provided, support experiences that feel fragmented, and sales teams reconstructing account history manually before every renewal, escalation or executive review.
Stateless systems cannot preserve organizational memory
The AI models themselves are becoming more capable, but the organizational foundation beneath them remains fragmented.
Most AI systems today are fundamentally stateless. They generate outputs based on temporary context windows rather than durable organizational memory. Every interaction requires the system to repeatedly reconstruct understanding from fragments.
Consider how databases work. We do not recompute everything from scratch every time a query arrives. We cache and index, then preserve relationships between entities, because continuously recomputing context is computationally irrational.
Yet much of enterprise AI is still being deployed exactly this way and the industry has started mistaking activity for intelligence.
What I believe organizations should focus on is whether they have structured, durable memory that lets AI and humans reason from the same shared context. Without that foundation, AI outputs remain generic.
Most collaboration systems multiply this problem in two ways. First, they encode knowledge into naming conventions and tribal memory – the kind that lives in channel names nobody can decode and folder structures only three people understand. New employees are not learning the business, they are learning the conventions.
Second, even when information exists, it remains inaccessible. The same decision appears as “PostgreSQL migration”, “database move Q3”, and “backend infrastructure change” across three different channels. They are semantically identical but textually invisible to any system trying to surface it.
This problem becomes even more acute in distributed organizations. I don’t believe you can build modern global companies on a “you had to be there” culture. Yet many businesses still operate as though important context naturally transfers through proximity and synchronous communication.
Search is not the same as understanding
Search was designed to discover information, whereas modern enterprise work requires systems that understand the relationships in data.
A customer escalation is not just a support ticket. It is connected to product decisions, engineering discussions, account history, contractual obligations, and revenue impact. A sales opportunity is tied to customer sentiment, historical support patterns, product usage, and internal stakeholder alignment.
Traditional collaboration systems flatten these relationships into disconnected channels and documents, whereas AI knowledge graphs preserve them.
Researchers call this a transactive memory system: the collective understanding of who knows what, how decisions were made, and how work is coordinated across teams. The same logic now extends to AI. Intelligent systems can participate in that process too by encoding context, surfacing relevant history, and routing knowledge to the right people at the right time.
Britain’s productivity problem is becoming an AI problem
The Office for National Statistics has consistently flagged weak productivity growth as one of the UK’s most persistent economic challenges. Since 2010, UK productivity has grown at 6.2%, compared with roughly 10% across the euro area and nearly 15% in the United States over the same period. AI is increasingly being positioned as a mechanism to help close that gap.
But productivity does not improve because your business has added more AI agents to the workflow. If every important decision still requires humans to manually reconstruct fragmented context, organizations just accelerate confusion.
What UK businesses need are systems capable of preserving context, maintaining institutional memory, and grounding AI systems in trusted organizational knowledge. Better AI infrastructure starts with a simple question: Does your organization remember anything? For most, the honest answer is no.
We’ve featured the best AI tool.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit
https://cdn.mos.cms.futurecdn.net/qP76MS2BAb7kSuWrvJXXYL-2560-80.jpg
Source link
