Fresh from the network

A large analysis from Germany finds no job losses due to AI, but first signs among young employees. What this means for planning the next generation of staff.

Application numbers are rising and many HR teams turn to AI tools that are not approved. What this means legally and what a sound solution looks like.

A tech chief calls for an emergency brake for AI, a test model wrecks its own environment. What companies should settle before their first productive agent.

A critical flaw in an SAP programming library shows that cloud customers, too, must actively maintain the extensions they have built themselves.

SAP is significantly expanding its range of AI agents. For companies running existing ERP systems, the key question is under which conditions they can use these functions.

A class action in the US accuses a restaurant chain of coordinating prices via an AI system. What companies should watch out for with pricing tools.

A European search engine replaced its language model twice within five months. What companies can learn from this for their own architecture.

A study of nine well-known AI services shows that all of them pass data to third parties. For companies, the way of access is what counts.

A major AI provider now marks machine-generated text covertly. Why a match says little about a text's quality and what companies should put in place.

The European Commission mainly wants to bring AI into the economy quickly instead of competing for the strongest models. What mid-sized companies can take from this.

At an Italian bank, an AI-generated voice led to a transfer of 95 million euros. Such fraud attempts are also rising in Switzerland.

Gartner expects many vendor-built AI agents to be switched off again. Companies that settle ownership, operation and exit early stay in control.

SAP sees AI agents as its biggest growth opportunity, OpenAI launches general always-on agents. For mid-sized firms, a different calculation counts.

A detailed single measurement shows that a task given to an AI agent uses many times the energy of a chat query. The reason lies in the context.

The German parliament is replacing private AI tools with its own secured offering. The approach transfers well to mid-sized businesses.

An unreported security incident and a new push on product liability show how open the question of responsibility for damage caused by AI models still is.

A vulnerability in four widely used programming assistants shows why code from AI tools needs the same checks as any other software.

Independent measurements show that a model with a low token price can cost more per task than a top model. What this means for budgeting.

Austria allows authorities to issue rulings fully automatically and ties this to clear conditions. They transfer well to business processes.

19 EU states are pooling more than ten billion euros for industrial AI. Switzerland is not part of it, yet its suppliers will still feel the effects.

SAP now only lets third-party AI agents reach its interfaces through its own channels. For companies, the question moves from technology into the contract.

A new survey shows that most companies now use AI, yet hardly any make full use of it. The money goes into data and connectivity, not licences.

A new policy paper challenges the view that strict data protection holds back innovation. The real problem, it says, is the uneven burden on smaller firms.

Agents that act on their own need a well designed approval step. Without it, adoption fails on governance or on acceptance.

Language models answer questions about company data. The answer only becomes reliable once metrics are defined and ownership is clear.

New tools automate data transfer during a system change. The expensive part of a migration stays untouched and becomes more visible as a result.

Stolen session keys drive up usage on consumption based AI services. Without itemised records, the abuse stays invisible.

Agent tasks can now run on fixed schedules. That removes the person who used to see every result.

Large employers now ask candidates about their AI practice. Adopting that question requires internal clearance first.

Voluntary AI test seals arrive from late 2026. They complement the statutory conformity assessment rather than replacing it.

New tools read thousands of contracts and turn them into tables. The value only appears once the result feeds back into a process.

After acquisitions, data sits in separate systems. A shared analytics layer helps, but it does not replace system consolidation.

From late 2027 some 3,000 federal staff will run an open-source suite alongside Microsoft 365. Why that is not an exit, and what companies can take from it.

Two incidents at major AI providers show the same pattern. An agent is limited by what its environment technically allows, not by what it was told.

A new global survey shows that nearly every company uses AI, while very few can quantify its contribution to profit. Here is why, and what follows from it.

More than a hundred technology companies jointly warn about AI-driven cyberattacks. The appeal targets critical infrastructure, yet smaller firms are especially exposed.

A large corporation halted its programme for an AI-native organisation shortly before announcing it. The published figures are instructive for any rollout.

A recent hands-on test shows what language models on your own hardware can do today. For companies handling confidential data, a second look pays off.

SAP shares gained roughly a third within a month. The more telling figure is the gap between booked cloud volume and the AI agents actually available to customers.

Market data, cloud momentum and embedded AI point in one direction: the core system of enterprise IT is not disappearing, it is gaining importance.

A major CRM vendor is opening its data and functions to an AI assistant. For ERP users, that turns one question into a selection criterion.

A flaw in a widely used AI assistant shows why permissions and connections become the real security question in day-to-day operations.

Artificial intelligence speeds up existing routines in both directions. Why clean data and clear responsibilities in the ERP decide the return.

Enterprise software specialists remain scarce and expensive. We show how a standard-based cloud architecture structurally reduces dependency on rare experts.

Digital assistants are taking over more and more business communication. We show what current figures and trust research mean and which rules prove effective.

AI now writes large parts of new software. We explain why system knowledge is becoming more important and which guardrails prove effective in the ERP context.

Since 2 August 2026, the EU AI Act requires transparency in AI use. We outline which duties now apply to operators of ERP systems with embedded AI functions.

Requirement catalogues generated and answered by AI are losing their value. What matters instead when mid-sized companies choose a new ERP system.

The old website is history: we have completely replaced our presence with a modern solution, vivid design, interactive content, an AI assistant and an editorial pipeline that delivers fresh insights.

Copilot Chat in Outlook now answers questions across the entire mailbox and calendar instead of just the open thread. And it needs no paid add-on.

Nineteen days without access to two leading models: the June 2026 incident shows what a serious AI strategy for mid-sized Swiss companies now has to cover.

Voice and text questions to our website assistant are now processed entirely in Swiss data centres, faster than before and without your question ever leaving the country.

What Copilot creates from confidential sources now automatically carries their protection level. Watermarks for generated video and audio come too. But are off by default.

Microsoft has introduced an approval process, lifecycle management and a licence requirement for security functions around self-built agents. Tinkering becomes operations.

Anyone asking a question on the website now reaches us directly, during the day a real person responds, around the clock our speaking assistant does.

The agentic workspace in Microsoft 365 has left free preview and has been billed by consumption since 1 July. Anyone who used the preview should now set spending limits.

Microsoft has expanded regional data processing for Copilot, and Copilot Chat now falls under the EU Data Boundary. One function remains outside it: web search.

In June 2026 two frontier models had to be switched off by official order. The incident shows where the real operational risk in using AI actually sits.

Since June, Copilot features in Word, Excel, PowerPoint and OneNote have been limited to licensed users. Since July there are business bundles with Copilot permanently included.

For ongoing conversations, we provide individual, access-protected pages: documents, points of contact and next steps all in one place.

Ready-made assistants, a knowledge graph as a guardrail and one common governance model: a sober look at what actually becomes usable from Q3 2026 onwards.

At the May 2026 conference the autonomous enterprise was declared. Behind the term sits more engineering than marketing and a question of governance.

A fast close is not a question of effort but of order. What it actually takes and why management benefits from it far more than accounting does.

Since April 2026 freely licensed language models have been available that run locally and work multilingually. For sensitive data that opens a serious alternative.

Processes and interfaces take centre stage. The lasting value arises somewhere else: in a continuous, machine-readable picture of the whole company.

Modern rollouts reverse the calculation: the standard stays untouched, while a large share of the funds flows into processes, training and people.

The shift to subscription models does not only affect user organisations. It fundamentally challenges the business model of many implementation partners.

The major cloud outage of October 2025 showed how tightly modern services are nested. The lesson is not to abandon the cloud, but to spread the risk better.

The digital workplace decides whether an ERP rollout is accepted. Why a role-based entry point achieves more than any additional specialist function.

Many still see their own hardware as the safest option. Practice suggests otherwise: location, access paths and a workable exit strategy are what count.

Results from language models depend less on the model than on how the task is phrased. Which prompting techniques hold up in ERP settings, and why documentation decides.

Announced a year ago as the foundation for data-driven decisions. Today we can check which connections actually work and what that means for mid-sized firms.