7 min read · 10 sources

Snowflake wants its AI agents to keep working after you log off. Its CoCo and CoWork Bots are designed to run ongoing tasks in the background, act through tools teams already use, and pull people in when a decision or sign-off is needed.

That’s the useful promise - and the operational question. An agent that acts while nobody’s watching needs clear permissions, reliable context, and a way to stop it before a bad assumption becomes a completed task. Snowflake’s announcements put those control problems alongside the pitch.

Snowflake’s new agents run in the background and only ping humans when a decision or sign‑off is required.

Snowflake’s agents are built to work between check-ins

Source: snowflake.com ↗

Snowflake CoCo and CoWork are in development as always-on agents for recurring work. Rather than waiting for an employee to ask a question, the bots are intended to work in the background and bring people in when work is ready or a human decision is required.

The plan is to fit into existing workflows, with automations in CoCo and CoWork and integrations with Slack, spreadsheets, and mobile devices. Shared skills, memory, and artifacts such as dashboards and reports are meant to let teams build on prior work. The practical change is a move from a chatbot that responds to a prompt toward software that can keep a task moving without one.

That makes the handoff the key engineering detail: what can an agent do on its own, what does it need to escalate, and how can a team inspect what it did? Snowflake’s release describes the intended behavior, but doesn’t provide details here on limits, reliability, or deployment timelines.

Refrigeration systems became a remote attack surface

Source: itpro.com ↗

More than 40 supermarkets across seven countries were affected by suspected cyberattacks that remotely raised refrigeration temperatures and spoiled food, according to ITPro’s report. Reports involved supermarket chains in Belgium, Luxembourg, Sweden, and Ireland, among others. The cause is still described as suspected; the incidents were reportedly detected after refrigeration equipment was tampered with remotely.

This is an operational technology problem, not just an IT account problem. Connected refrigeration controls can have direct physical consequences, while equipment may stay in service long after software support ends. A command path that lets a remote system alter temperatures needs authentication, network boundaries, and limits on what values it will accept.

The reported scale - hundreds of thousands of euros’ worth of food spoiled - makes the cost of leaving connected equipment exposed tangible. Operators need to know which devices are internet-accessible, who can issue commands, and whether controls can reject dangerous instructions regardless of where they came from.

AI users may be working longer, not less

Source: cio.com ↗

ActivTrak’s latest research found that employees who use AI most heavily work up to 90 minutes longer per day, switch apps more often, and have shorter focus sessions, according to CIO. That clashes with the company’s earlier research, which found employees were working less overall.

The reports aren’t necessarily measuring the same thing: the sample sizes and collection periods differ, and time spent using an AI tool doesn’t establish whether the work it supports is faster or better. More app switching could mean that AI adds coordination and review work, or simply that the heaviest users have different jobs. The available findings don’t settle the question.

For teams evaluating AI, the important distinction is between activity and outcomes. Tool usage, hours worked, and app switching are signals, not a productivity score. Measuring whether a workflow got faster or produced better results requires looking at the work itself.

Gartner warns AI-led layoffs may be expensive to reverse

Source: itpro.com ↗

Gartner predicts that 30% of employees laid off and replaced by AI will be rehired by 2029, potentially at a higher cost, according to ITPro. The firm says companies risk making cuts before they understand how work will change.

The planning gap is substantial: only 19% of firms surveyed said they have practices to help leaders assess AI’s impact on work and align talent plans with business goals. Automating a task doesn’t automatically remove the surrounding work: people may still need to validate outputs, handle exceptions, or maintain the process.

For engineering and IT leaders, this is a reminder to map workflows before treating a model’s capability as a staffing plan. Rehiring costs can follow from removing the people who understood the edge cases.

Snowflake puts agent controls and business definitions in one place

Source: snowflake.com ↗

Snowflake says its generally available Cortex AI Gateway centralizes controls for agent connections, model access, and spending. The gateway extends governance to agents, giving organizations visibility into usage and a place to manage policies.

A separate Business Ontology feature is planned for preview. It’s intended to give agents approved definitions of business terms, so that “customer” or “revenue” has a governed meaning rather than one inferred differently across data sources. That context matters when agents produce answers or take actions based on company data.

Sandbox execution and agent-abuse detection are still in private preview, according to the newsletter. That distinction matters for teams assessing what they can deploy now versus what remains in testing. The architecture is aimed at central oversight, but it doesn’t remove the need to define who owns each agent and what it’s allowed to do.

Bull and Schwarz Digits plan a European AI stack

Source: itpro.com ↗

Bull and Schwarz Digits signed an agreement to develop a sovereign AI architecture for European organizations in regulated and strategic sectors, ITPro reports. Schwarz Digits brings its STACKIT cloud, while Bull is contributing computing hardware, liquid cooling, and AI software.

“Sovereign” infrastructure is meant to give customers more control over where workloads and data run, and which providers operate them. That can matter for regulated organizations, but the announcement is a plan to collaborate, not a detailed service specification. Buyers will still need to assess the resulting system’s workload support, availability, and operating model.

Teesside’s data center plans come with a power timeline

Source: itpro.com ↗

National Grid plans a new substation at Dorman Point in Teesside to support data centers, battery storage, and other power-intensive industry, according to ITPro. The company is buying 73.3 acres, with the site located beneath 275kV and 400kV transmission lines.

The substation is expected to start with 1,440 MVA of capacity and come online by the end of 2029. That schedule puts a physical limit on how quickly new power-hungry projects can move from proposal to operation. Data center plans depend on grid infrastructure that takes years to build, not just on securing land and servers.

CIOs are deploying AI faster than they can operationalize it

Source: accenture.com ↗

Accenture’s survey of 300 technology leaders across 19 industries and 16 countries found that 64% are scaling enterprise AI deployment, but only 7% are optimizing it at scale, according to its CIO Pulse report. User adoption and training are the most frequently cited barriers. Nearly half still require manual approvals for agents in IT operations.

The gap is between putting AI into workflows and getting measurable value from it. Manual approval can be a sensible safety control, but it also limits how far an agent can operate without human intervention. The survey suggests organizations are still working out where to automate, what to measure, and which controls need to stay in the loop.

UK businesses have less than 100 days to move off copper

Source: itpro.com ↗

Openreach warned that roughly 300,000 business copper lines remain active ahead of the UK’s PSTN shutdown on January 31, according to ITPro. Lines may support payment terminals, alarms, or building systems, not just desk phones.

Any service still reliant on copper risks disruption after the deadline, and Openreach says there won’t be an extension. Businesses need to inventory what’s connected before swapping phone services; replacing a handset won’t migrate an alarm or a building system that also depends on the line.

IAM spending is forecast to grow 55% by 2031

Source: itpro.com ↗

Juniper Research expects global identity and access management spending to rise from $26.2 billion in 2026 to $40.5 billion by 2031, a 55% increase, ITPro reports. The forecast reflects more organizations managing AI agents and other nonhuman identities.

The shift is from authenticating a person once to continuously assessing access as context changes. Agents add a different identity-management problem: they need owners, scoped permissions, and a way to revoke access when a task or service ends. As more software acts on behalf of employees, IAM policies designed only around people will need to catch up.