The Agentic Stack Gets Secured

What happened:
Palo Alto Networks announced plans to acquire AI-security startup Koi. The company warned that AI agents and plug-ins create a new attack surface, behaving like “ultimate insiders” with legitimate credentials and deep data access. Koi’s tech provides visibility and control over agents, scripts, model artifacts, and plug-ins. Palo Alto will fold it into its Prisma AIRS platform for model scanning and governance.

Why it matters:
As AI agents embed into workflows, traditional endpoint security isn’t enough. Autonomous systems can move data, call tools, and operate outside classic controls. Agent-native security is becoming table stakes.

What’s next:
Expect cybersecurity vendors to race toward AI-specific detection, observability, and agent governance tools.

What happened:
An IT Brew guide urged IT teams to slow down before going “fully agentic.” Experts recommend starting with risk assessments, workload analysis, and a clear business case—rather than defaulting to autonomy because it’s trendy.

Why it matters:
Agentic AI is powerful—but misapplied autonomy can create chaos. Smart deployments begin with governance and clear ROI, not hype.

What’s next:
Enterprises will increasingly adopt “autonomy audits” before rolling out multi-step AI agents.

What happened:
Workflow platform Temporal raised $300 million to build a “durable execution” layer for agentic AI. The company argues that most AI agents fail because underlying systems can’t manage long-running, stateful workflows. Customers include OpenAI and ADP.

Why it matters:
Autonomous agents need memory, retries, and persistence. Without execution infrastructure, agents stall, duplicate tasks, or lose context.

What’s next:
The “agent reliability stack” may become as important as the model itself.

What happened:
Microsoft’s Worklab predicted a shift from co-pilot tools to full “systems of work.” Planner agents will design workflows; worker agents will execute and verify tasks end-to-end.

Why it matters:
AI isn’t just assisting anymore—it’s orchestrating. That requires redesigning business processes around autonomous execution.

What’s next:
Expect enterprises to restructure teams around oversight and coordination rather than task execution.

What happened:
Analysts argue Nvidia’s next-gen GPU systems (like NVL72) could slash inference costs and enable real-time agentic AI at scale. GPU-first architectures may fundamentally shift SaaS and cybersecurity economics.

Why it matters:
Cheaper tokens = more autonomy. AI-native vendors can run complex agents affordably, putting pressure on legacy SaaS models.

What’s next:
The AI cost curve could determine which vendors survive the shift to agentic software.

What happened:
The U.S. Department of Defense reportedly wanted unrestricted access to Claude models for lawful uses, including potential weapons and surveillance. Anthropic declined, citing ethical policies.

Why it matters:
This is the front line of AI governance: model providers imposing limits vs. governments seeking strategic leverage.

What’s next:
Expect deeper debates over who controls advanced AI capabilities—and under what constraints.

What happened:
• Meridian raised $17M for an IDE-style financial modelling workspace.
• Vega Security secured $120M for AI-native cloud security.
• SiFi raised $20M for AI finance automation.
• Infosys partnered with Anthropic to embed Claude across enterprise workflows.
• Cohere launched Tiny Aya, multilingual edge models (70+ languages).
• Adani committed $100B to renewable AI data centers in India.

Why it matters:
Capital is concentrating around infrastructure, finance automation, and sovereign AI ecosystems. The agent economy is global—and scaling fast.

What’s next:
Watch for more edge-friendly models and national AI infrastructure bets.

What happened:
Financial and security analysts called for deterministic controls, continuous observability, and better identity management for AI agents. As autonomy rises, so does compliance risk.

Why it matters:
Autonomous systems behave probabilistically—but regulators don’t. Governance must evolve before agents handle sensitive workflows.

What’s next:
Non-human identity management may become a new security category.

What happened:
San Francisco Fed President Mary Daly said policymakers are evaluating AI’s potential productivity impact. Faster growth could influence inflation and interest-rate decisions.

Why it matters:
If AI boosts productivity meaningfully, it could shift economic baselines and monetary policy.

What’s next:
Expect central banks to factor AI into long-term growth models.

What happened:
Ericsson launched an Agentic rApp-as-a-Service on AWS, enabling network teams to optimise systems using natural language. Early trials show operational gains.

Why it matters:
This is agentic AI moving into mission-critical infrastructure. Operators can now “talk” to networks instead of scripting manual optimisations.

What’s next:
Telecom may become one of the first industrial sectors fully redesigned around agent orchestration.