Agent Architecture and Memory Engineering

• Seven layer architecture becomes a shared build reference
• Five layer memory model targets agent amnesia
• Retrieval, tools, agents, memory each fix one LLM limit
• Ontology plus knowledge graph beats a semantic layer alone
• TypeSafe AI's Jev returns typed decisions across 30 use cases
• Speculative decoding triples token speed on NVIDIA Jetson
• Eight model types are combining, not consolidating into one

‍

Agentic AI Security and Containment

• OpenAI's test model bypassed a scanner, ignored instructions
• Agent moved laterally during a Hugging Face evaluation
• Gemini hacked three real companies in a May test
• Unreleased OpenAI model planted instructions in its own summaries
• Splunk SOAR proposed as an automated circuit breaker
• Agents run disposable inside hardened Kubernetes sandboxes
• Confidence and authorization split into separate gates

‍

Agent Identity, Governance Tooling and Data Trust

• Entra Agent ID extends Zero Trust to agents
• Windows 365 for Agents isolates each agent on its own Cloud PC
• Security focus shifts from models to agents themselves
• New architecture packages 36 governance controls for MSSP teams
• Reference architecture offers a 12 month governance roadmap
• Shadow AI featured in 43 percent of breaches
• Oracle AI Database ties agentic scaling to source data security

‍

Regulation and Policy

• California now runs nineteen AI laws, ten binding in 2026
• OpenAI's shutdown work echoes an EU AI Act requirement
• Every major AI model now watermarks generated text
• Use-case governance push points to EU AI Act Annex III
• New Senate bill creates a custodial user agent category
• Apple withheld Siri AI features in the EU

‍

AI Governance and Risk Programs

• Governance pitches describe features, not proof of action
• Circle of control separates governable risk from geopolitics
• Sign-off steps on clinical workflows produce checkboxes, not oversight
• Three governance tracks move in parallel, not convergence
• Program built on "what not to build" wins a DataIQ Award

‍

Enterprise Economics, Adoption and ROI

• Agent workflows burned budgets despite falling token prices
• Business cases should separate today's ROI from future bets
• Splunk Tokenomics pairs token cost with agent performance
• Vera Rubin NVL72 claims up to 35x lower cost per token
• Agentic value comes from retired steps, not seats deployed
• Disagreeing systems create a reconciliation economy behind agents
• Deel's Akai learned a workflow from one screen recording

‍

Workforce, Leadership and Human Oversight

• Spec writing and critical reading are now top developer skills
• One firm gated new-hire AI access behind proven judgment
• Manual gates proposed to stop human skills decaying
• Unclear AI accountability read as discomfort, not confusion
• Real leadership risk is AI's effect on work, not doom
• Ten agent content team lost 43 percent of time to rereading
• Trust in agentic AI lags well behind adoption

‍

Learning Resources, Books and Community

• Thirteen NVIDIA certifications map to 2026 upskilling paths
• Ten free resources build AI product management skills
• HR agentic AI course passes 40,000 learners
• New interactive book tracks a repo assistant's growth
• Two new books land on AI security and vibe coding
• One event built 50 AI skills in 24 hours
• Eleven startups at Llama Lounge show AI going vertical

‍

Thanks to Ashish C Singh, Alistair Greenwood, Shivani Virdi, Guru Bandiatmakur, Greg Coquillo, Shrey Shah, Chen Su, Jean Malaquias, Rudy Bakalov, Sujit Chandrapati, Eva Benn, André Cardoso, Sol Warnock, Ajay Punreddy, Rajesh Iyer, Frank Gundlich, Travis Guinn, Aiswarya Venkitesh, Stefan Kinnestrand, Karthik Chakravarthy, Anthony Fuller, Joseph Davis, Ana Carolina Bonaldi Cayres Pinto, Michael Pan, Joseph Abraham, Krystian Bień, Trent Cotton, Aleksandr Tiulkanov, Nikhil Kassetty, Christina Boesenberg, Prof. Hernan Huwyler, Colin Cosgrove, Sigrid Berge van Rooijen, Jean Gan, Stephanie Gradwell, Ameena Syeda, Pradeep Sanyal, Oliver Tuszik, Smriti Mishra, Seow Yin Lim, Dr. A. Denise Holtschulte, Joanna Miler, Gabriel Millien, Andreas Horn, Tiffany Masson, Richard Self, Joshua Miller, FAISAL HOQUE, Jimmy Acton, Spyridon Georgiadis, Vishakha Sadhwani, Shania Rasheed Nalagath, Terri Horton, Maxim Salnikov, Amit Singh, Yacob Cohen Arazi, Bora Ger, Philippe Ensarguet and Silicon Valerie Bertele for the insights that shaped this edition. Find the full list of posts and voices on LinkedIn: