The AI market is moving from individual chat interfaces towards governed agent systems, infrastructure-heavy model development and more complicated deployment risk. This week's curated links highlight the operating models, compute pressures, security questions, and cultural signals that project leaders should keep an eye on.
Enterprise Agents and Operating Models
Anthropic launches agents for financial services: Anthropic released ten prebuilt finance agents for workflows including pitchbooks, KYC review, earnings analysis, credit memos and month-end close. The important signal is the packaging of agents around repeatable, auditable business processes.
IBM sets out an AI operating model at Think 2026: IBM is framing the next stage of enterprise AI around operating models rather than isolated tools. The piece is useful for leaders thinking about how pilots become managed business change.
IBM Consulting expands AI capabilities: IBM Consulting is expanding AI transformation services as demand shifts from experimentation to implementation. It points to the growing market for advisory, governance and integration support around AI programmes.
KPMG explains enterprise AI agent strategy: KPMG sets out how organisations should think about agent strategy, human oversight and control frameworks. The practical value is in treating agents as part of the operating model rather than a standalone software purchase.
Salesforce outlines AI agent trends: Salesforce focuses on identity, governance, handoffs and measurable outcomes for enterprise agents. It is a useful read for teams preparing to manage agent fleets across customer, sales and operational workflows.
Models, Compute and Long Context
DeepSeek could reach a $45 billion valuation: DeepSeek is reportedly heading towards a major first investment round that could value the company at $45 billion. The valuation reflects the continued importance of lower-cost frontier model challengers in the global AI market.
Claude is tested as a Meta ads expert: This evaluation of Claude as a specialist advertising operator is a useful reminder that teams need workflow-specific model tests. General model rankings matter less than whether a system can produce reliable outputs for a defined task.
Harvard study finds AI improved emergency diagnosis accuracy: A Harvard study found AI improved diagnostic accuracy in an emergency medicine setting. The broader lesson is that structured, high-stakes review tasks remain one of the strongest areas for AI augmentation.
Governance, Risk and Regulation
Public Citizen launches a state chatbot legislation tracker: Public Citizen has launched a tracker for US state legislation regulating chatbots. The key point is that AI regulation is becoming more fragmented and more operationally relevant for firms working across multiple jurisdictions.
Opaque summarises international AI agent security guidance: Opaque summarises guidance from agencies across five governments on securing AI agents. The consistent message is that agents need identity, permissions, monitoring and containment before they are connected to sensitive systems.
BBC covers AI chatbot delusions: The BBC examines cases where chatbot interactions appear to have contributed to delusional thinking. It is a reminder that responsible AI policies need to address user behaviour and dependency risks alongside data security.
Chinese courts limit replacing workers with AI: Chinese courts have ruled that companies cannot dismiss workers simply to replace them with AI. Employment law is starting to shape how automation can be deployed in the workplace.
Paul Boudreau argues that blocking AI at work does not work: Boudreau argues that blocking workplace AI use usually pushes people into unmanaged channels. The stronger path is governed access, training and clear workflow design.
Robotics, Culture and Behaviour
Figure shows humanoid robots cleaning restaurant spaces: Figure's restaurant-cleaning demonstration is another signal that humanoid robots are being tested against practical service tasks. The useful question is how reliability, supervision and unit economics develop as deployments move beyond demos.
When Dawkins met Claude: This cultural read shows how high-profile thinkers are testing frontier models as reasoning partners. It helps explain why model adoption is becoming behavioural as much as technical.
Scott Galloway on AI and economic concentration: Galloway discusses AI, labour and economic concentration on The Diary of a CEO. The episode is useful for thinking about who captures AI value as automation becomes more embedded in work.
We’ll carry on our exploration for more curated news for our next edition… Until then, happy reading!
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
