News

• Home puts chats, delegated work, and Office documents inside one interface
• Code lets non-programmers describe software and have Copilot build it
• Autopilot can continue recurring work without waiting for another instruction

Microsoft has introduced a redesigned version of its Copilot AI platform which claims to combine chat, delegated work, and coding tools into a unified application experience for users.

The company says the update is meant to let individuals and organizations scale artificial intelligence across everyday tasks and long-term projects.

Three new capabilities anchor this release, including Home, Code and Autopilot, each aimed at a different kind of work.

The Home tool brings together two existing modes, Chat for quick questions and Cowork for tasks users delegate entirely, under one shared starting point.

Word, Excel, and PowerPoint now operate inside this same interface, letting users draft documents, budgets, and presentations without switching applications.

In these documents, Copilot is now grounded in Fabric IQ, pulling context from more than 20 million semantic models built in Power BI.

Edits made by colleagues or by the assistant itself appear in real time, so progress stays synchronized across a shared file.

A new plugin registry will let organizations manage Microsoft, partner and custom-built plugins from one central catalog starting this month.

Independent developers and partners can also publish plugins once for use across multiple Copilot surfaces under the new registry arrangement.

The Code tool allows non-programmers to describe an app, tracker or dashboard in plain language and have it built automatically.

This feature runs on the same underlying technology used in GitHub Copilot and can be hosted within a company's own systems.

Autopilot, the third addition, is a persistent agent capable of completing recurring work without needing a new instruction each time.

It can run supplier reviews or similar multi-step processes, build schedules, contact stakeholders, and follow up on outstanding items independently.

Because it operates continuously in the cloud, work can continue late at night or whenever a person's attention shifts elsewhere.

A related feature called Today, entering private preview in October, will summarize missed messages and pending tasks across mail and chat.

Microsoft is also tying spending controls to these tools through a system it calls FinOps for AI, letting administrators track usage.

Administrators can set spending limits, approve credit requests and restrict which AI models different teams are permitted to use each month.

Everyday tasks like quick answers or first drafts run on a fixed-price subscription, while agentic features use usage-based billing.

Code, Cowork and Autopilot all fall under this usage-based pricing model, alongside frontier models Microsoft refers to as Astra and Fable.

Home and Code are set to roll out through Microsoft's Frontier program within weeks, and Autopilot enters private preview by month's end.

Microsoft has not released independent data showing how widely the three features are being adopted, how accurate they are, or how much time they actually save.

• £90 million is being allocated toward using excess heat from UK data centers
• In London, around 650,000 homes will receive cut-price heating via a “mega heat project”
• The project is described as delivering “clean heat” to London

The UK government plans to heat around 750,000 homes as part of a new £90 million project that will include supplying hot water heated by data centers. The Department for Energy Security and Net Zero (DESNZ) is funding the building of new heat networks and improving existing infrastructure to pipe hot water from data centers across parts of London, including under the Thames.

The initiative is part of a national project to reduce heating bills, with London, Manchester, Newport and Abergavenny in Wales, Solihull in the Midlands, and Bradford in Yorkshire all benefitting. Similar waste-heat projects have been running in Copenhagen, Amsterdam, Vienna, and Berlin for several decades.

In London alone, 650,000 homes will benefit, although the project has drawn ridicule for the cited use of “electric barges” to transport hot water 25 kilometers along the River Thames, while the subterranean pipe system is installed.

Diverting heat from data centers into water is the core of the £90 million in government funding, but the mega heat project in London relies on unlocking £5 billion in investment, which will also generate jobs.

Alongside roles in river transport, the money will support construction and engineering projects. The mega heat project under the Thames is expected to distribute heated water from a data center, where enough excess heat is generated to provide enough for homes across London.

While £41 million of the funding goes to the capital, £9.7 million will be invested in extending the city’s original heat network. Heat networks were common in the industrial north in the mid-20th century, although Bradford’s network is a more recent project, commencing in 2024. The projects in Greater Manchester (Gorton and Levenshulme) will replace legacy heating systems in social housing blocks.

While initial reaction to the news of London's mega heat project has been positive, the use of “electric barges” to transport water along the River Thames has inspired some derision. However, the £41 million heat project is planned to start as soon as possible.

The barges (which will cover 25 kilometers a day) are heated, with the water temperature sustained until it is deposited into insulated riverside tanks mounted. Expected to launch in 2028, the barges will remain in operation until the new hot water distribution system is completed.

“Clean power and clean heat benefits communities across our country," noted the UK’s Minister for Local Energy and Jobs Martin McCluskey.

"Building the infrastructure to deliver it is reindustrialising the UK and rolling out cutting-edge energy technology so families can benefit from lower bills, good jobs and energy security. And for those already using heat networks, we are upgrading inefficient systems to make sure people get the high-quality, low-cost, reliable heating they deserve.”

For the past several years, the enterprise AI conversation has been dominated by models. Every week has seemed to bring a new benchmark, a new leader board, or a new debate about which model is smartest. Organizations often made technology decisions based on incremental improvements in model performance, assuming that the model itself was the primary source of competitive advantage, but that assumption is rapidly becoming outdated.

As foundation models continue to improve, they are also converging in capability. At the same time, the cost of deploying AI at scale is becoming a more pressing concern for enterprises than squeezing out marginal gains in benchmark performance.

As a result, the focus is shifting from the model itself to everything surrounding it: the systems that make AI useful, governable, and economically viable in real-world business environments.

The question enterprises are increasingly asking is not, "Which model should we use?" Instead, they are asking, "How do we turn AI into reliable work?"

The answer begins with what can be described as the harness. An AI model on its own is powerful, but incomplete. To deliver business value, it needs context, access to enterprise tools and data, memory, controls, and guardrails. Together, these elements form the scaffolding that transforms a general-purpose model into a system capable of performing useful work, determining how effective an AI system becomes.

Two organizations might deploy the same underlying model yet achieve dramatically different results. One may struggle with inconsistency, high costs, and poor outcomes. The other may produce reliable, measurable business value. The difference often lies in the quality of the harness surrounding the model rather than the model itself.

That reality becomes more significant as models continue to commoditize. In many cases, organizations will have access to the same leading AI capabilities. What will differentiate them will be their ability to operationalize those capabilities efficiently and safely, and move from intelligence creation to intelligence application.

A related concept is what might be called the loop. Many organizations still approach AI interactions as isolated prompts and responses. But enterprise work rarely operates that way. Real business processes involve objectives, verification, correction, and completion criteria. A loop defines a governed unit of work.

Rather than simply generating an answer, the system performs a task, validates progress against the objective, corrects mistakes when necessary, and stops when the outcome is achieved. This seemingly subtle distinction has enormous implications.

When AI systems operate through loops instead of one-off prompts, they generate operational traces and performance records. Those traces can be used to evaluate outcomes, improve workflows, and refine future execution. Over time, this creates a learning system that grows more effective through use rather than requiring constant manual intervention.

In practice, this means enterprises are no longer deploying isolated AI tools. They are building repeatable learning systems — ones that improve the more they do the work.

As AI adoption expands, organizations encounter another reality: there is rarely a single harness. Most enterprises operate across multiple functions, business units, and regulatory environments. Each domain may require different tools, data sources, workflows, and governance requirements.

A harness optimized for software development is unlikely to be the same one used for finance, healthcare, customer service, or compliance. This creates a new challenge: orchestration.

The orchestration layer acts as the traffic controller for enterprise AI. It routes tasks to the right harness, determines when human oversight is required, and coordinates work across multiple systems. Rather than treating AI as a standalone capability, orchestration allows organizations to manage AI as an enterprise-wide operational platform.

This capability becomes particularly important in regulated industries, where workflows often cross organizational boundaries and require strong accountability. As AI moves deeper into critical operations, orchestration will become a defining architectural requirement rather than a technical afterthought.

If orchestration determines where work goes, governance determines whether that work can happen safely and assurance determines whether we can trust it was done as intended. Many organizations initially viewed AI governance and assurance as a compliance exercise rather than a core operational requirement.

Effective governance and assurance encompasses policy enforcement, authorization controls, asset management, cost monitoring, evaluation and audit services, observability, guardrails, and risk management. These capabilities must function consistently whether work occurs within a single AI system or across multiple interconnected workflows.

The need is especially acute as enterprises deploy autonomous and semi-autonomous systems. A model generating an inaccurate answer may be inconvenient. An AI agent taking action in a regulated business process without adequate oversight can be more problematic. Organizations that treat governance and assurance as an integral component of their AI architecture rather than a bolt-on feature will be better positioned to scale adoption while maintaining trust.

Collectively, the harness, orchestration capabilities, and governance and assurance layers form what can be viewed as a governed operating environment for AI. Combined with models, data, and compute infrastructure, they become an ecosystem that functions as a single accountable system.

This shift in perspective is important because no single vendor currently delivers every component of that ecosystem. The future of enterprise AI is therefore unlikely to be dominated by one model provider or one platform vendor. Instead, success will depend on how effectively organizations integrate and operate a diverse collection of technologies.

For enterprises, particularly those operating under regulatory, security, or sovereignty requirements, this integration challenge is not optional. AI must operate within boundaries defined by business policy and organizational control. The ability to combine frontier models with protected enterprise data while maintaining accountability will become a critical capability.

The model layer will continue to improve. It will also become increasingly interchangeable, but that does not diminish the importance of AI innovation. Instead, it changes where value is created.

The next battleground is the layer surrounding the model: the harnesses that make work reliable, the loops that make systems learn, the orchestration platforms that coordinate activity, and the governance and assurance frameworks that make scale enterprise ready. Together, they will determine whether AI remains an impressive technology demonstration or becomes a sustainable enterprise capability.

In the years ahead, the organizations that gain lasting advantage are likely to be those that stop asking which model is best and start focusing on how to integrate, govern, and operate AI as an accountable business system. The future of enterprise AI belongs not just to the model alone, but to everything around it.

We've featured the best AI website builder.

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